<!DOCTYPE html>
<html lang="en">
  <head prefix="og: http://ogp.me/ns# fb: http://ogp.me/ns/fb# 2490221586: http://ogp.me/ns/fb/2490221586#">
    <meta charset="utf-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1, user-scalable=no" />
    <meta name="include_mode" content="async">

    <!-- SL:start:notranslate -->
    <title>TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会)</title>
    <meta name="description" content="TensorFlow の基本的な使い方やコーディング方法を、線形重回帰やロジスティック回帰、パーセプトロンの実装例を通じて解説しています。（機械学習初心者向け） ※2016/02/27に開催した「GDG京都 機械学習勉強会」で発表した資料です。">
    <!-- SL:end:notranslate -->

    <meta name="robots" content="index">
    <meta id='globalTrackingUrl' content="https://www.linkedin.com/li/track">
      
  <!-- SL:start:notranslate -->
  <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  <meta http-equiv="x-dns-prefetch-control" content="on">
  <meta name="thumbnail" content="http://cdn.slidesharecdn.com/ss_thumbnails/machinelearningwithtensorflow20160227-160227125739-thumbnail.jpg?cb=1456578209" />
  <!-- SL:end:notranslate -->

    <link rel="shortcut icon" href="http://public.slidesharecdn.com/b/favicon.ico?d8e2a4ed15">
    <link rel="alternate" type="application/rss+xml" title="RSS" href="http://www.slideshare.net/rss/latest" />
    <link rel="search" type="application/opensearchdescription+xml" href="/opensearch.xml" title="SlideShare Search">

    <link href="http://public.slidesharecdn.com/b/ss_foundation/stylesheets/app_critical.css?770ef086df" media="screen" rel="stylesheet" type="text/css" />
    <!--[if IE 9]><link href="http://public.slidesharecdn.com/b/ss_foundation/stylesheets/ie9_nav_bar_fix.css?8fb8af5274" media="screen" rel="stylesheet" type="text/css" /><![endif]-->
    <link rel="alternate" hreflang="en" href="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg"/>
<link rel="alternate" hreflang="es" href="http://es.slideshare.net/ToruUenoyama/tensorflow-gdg"/>
<link rel="alternate" hreflang="fr" href="http://fr.slideshare.net/ToruUenoyama/tensorflow-gdg"/>
<link rel="alternate" hreflang="de" href="http://de.slideshare.net/ToruUenoyama/tensorflow-gdg"/>
<link rel="alternate" hreflang="pt" href="http://pt.slideshare.net/ToruUenoyama/tensorflow-gdg"/>
<link rel="alternate" hreflang="x-default" href= ""/>
    <script type="text/javascript">
//<![CDATA[
var smtId = "ab5bb963d";
var smtDefaultStyles = false;
var smtRedirect = true;
var smtProt = (("https:" == document.location.protocol) ? "https://" : "http://");

var smtPreRender = function(data) {
  for (i in data) {
    if (data[i].code === "en-us") {
      data[i].name = "English";
    }
  }
};

var smtRedirectMapper = function(locale, sites) {
  if (/^es/i.test(locale)){ //disable redirect to spanish site
    return null;
  }
  if (locale in sites) {
    return sites[locale];
  }
  if (/^fr/i.test(locale)) {
    return sites['fr-fr'] || null;
  }
  if (/^de/i.test(locale)) {
    return sites['de-de'] || null;
  }
  return null;
};

var smtElmt = document.createElement('script');
smtElmt.type = "text/javascript";
smtElmt.async = true;
smtElmt.src = smtProt + "cdn01.smartling.com/ls/"+ smtId +".js";
script = document.getElementsByTagName("script")[0];
script.parentNode.insertBefore(smtElmt, script);

//]]>
</script>

      <meta content="authenticity_token" name="csrf-param" />
<meta content="u0bE141K9xfzN1TYtg9EFsbuV0fX5Pxr0utESslAKIY=" name="csrf-token" />

      <meta content="index" name="robots" />

  <link href="http://public.slidesharecdn.com/b/ss_foundation/stylesheets/slideview_critical.css?390616d5d7" media="screen" rel="stylesheet" type="text/css" />

    

<link href="http://public.slidesharecdn.com/b/stylesheets/ssplayer/combined_presentation.css?b24096a132" media="screen" rel="stylesheet" type="text/css" />



    <link rel="dns-prefetch" href="//www.slideshare.net">
    <link rel="dns-prefetch" href="//public.slidesharecdn.com">
    <link rel="dns-prefetch" href="//image.slidesharecdn.com">
    <link rel="dns-prefetch" href="//cdn.slidesharecdn.com">
      <link rel="dns-prefetch" href="//cdn01.smartling.com">
    <link rel="dns-prefetch" href="//www.linkedin.com">

    <script type='text/javascript'>
      // Inlining Modernizr
      window.Modernizr=function(a,b,c){function d(a){t.cssText=a}function e(a,b){return d(x.join(a+";")+(b||""))}function f(a,b){return typeof a===b}function g(a,b){return!!~(""+a).indexOf(b)}function h(a,b){for(var d in a){var e=a[d];if(!g(e,"-")&&t[e]!==c)return"pfx"==b?e:!0}return!1}function i(a,b,d){for(var e in a){var g=b[a[e]];if(g!==c)return d===!1?a[e]:f(g,"function")?g.bind(d||b):g}return!1}function j(a,b,c){var d=a.charAt(0).toUpperCase()+a.slice(1),e=(a+" "+z.join(d+" ")+d).split(" ");return f(b,"string")||f(b,"undefined")?h(e,b):(e=(a+" "+A.join(d+" ")+d).split(" "),i(e,b,c))}function k(){o.input=function(c){for(var d=0,e=c.length;e>d;d++)E[c[d]]=!!(c[d]in u);return E.list&&(E.list=!(!b.createElement("datalist")||!a.HTMLDataListElement)),E}("autocomplete autofocus list placeholder max min multiple pattern required step".split(" ")),o.inputtypes=function(a){for(var d,e,f,g=0,h=a.length;h>g;g++)u.setAttribute("type",e=a[g]),d="text"!==u.type,d&&(u.value=v,u.style.cssText="position:absolute;visibility:hidden;",/^range$/.test(e)&&u.style.WebkitAppearance!==c?(q.appendChild(u),f=b.defaultView,d=f.getComputedStyle&&"textfield"!==f.getComputedStyle(u,null).WebkitAppearance&&0!==u.offsetHeight,q.removeChild(u)):/^(search|tel)$/.test(e)||(d=/^(url|email)$/.test(e)?u.checkValidity&&u.checkValidity()===!1:u.value!=v)),D[a[g]]=!!d;return D}("search tel url email datetime date month week time datetime-local number range color".split(" "))}var l,m,n="2.8.2",o={},p=!0,q=b.documentElement,r="modernizr",s=b.createElement(r),t=s.style,u=b.createElement("input"),v=":)",w={}.toString,x=" -webkit- -moz- -o- -ms- ".split(" "),y="Webkit Moz O ms",z=y.split(" "),A=y.toLowerCase().split(" "),B={svg:"http://www.w3.org/2000/svg"},C={},D={},E={},F=[],G=F.slice,H=function(a,c,d,e){var f,g,h,i,j=b.createElement("div"),k=b.body,l=k||b.createElement("body");if(parseInt(d,10))for(;d--;)h=b.createElement("div"),h.id=e?e[d]:r+(d+1),j.appendChild(h);return f=["&#173;",'<style id="s',r,'">',a,"</style>"].join(""),j.id=r,(k?j:l).innerHTML+=f,l.appendChild(j),k||(l.style.background="",l.style.overflow="hidden",i=q.style.overflow,q.style.overflow="hidden",q.appendChild(l)),g=c(j,a),k?j.parentNode.removeChild(j):(l.parentNode.removeChild(l),q.style.overflow=i),!!g},I=function(b){var c=a.matchMedia||a.msMatchMedia;if(c)return c(b)&&c(b).matches||!1;var d;return H("@media "+b+" { #"+r+" { position: absolute; } }",function(b){d="absolute"==(a.getComputedStyle?getComputedStyle(b,null):b.currentStyle).position}),d},J=function(){function a(a,e){e=e||b.createElement(d[a]||"div"),a="on"+a;var g=a in e;return g||(e.setAttribute||(e=b.createElement("div")),e.setAttribute&&e.removeAttribute&&(e.setAttribute(a,""),g=f(e[a],"function"),f(e[a],"undefined")||(e[a]=c),e.removeAttribute(a))),e=null,g}var d={select:"input",change:"input",submit:"form",reset:"form",error:"img",load:"img",abort:"img"};return a}(),K={}.hasOwnProperty;m=f(K,"undefined")||f(K.call,"undefined")?function(a,b){return b in a&&f(a.constructor.prototype[b],"undefined")}:function(a,b){return K.call(a,b)},Function.prototype.bind||(Function.prototype.bind=function(a){var b=this;if("function"!=typeof b)throw new TypeError;var c=G.call(arguments,1),d=function(){if(this instanceof d){var e=function(){};e.prototype=b.prototype;var f=new e,g=b.apply(f,c.concat(G.call(arguments)));return Object(g)===g?g:f}return b.apply(a,c.concat(G.call(arguments)))};return d}),C.flexbox=function(){return j("flexWrap")},C.flexboxlegacy=function(){return j("boxDirection")},C.canvas=function(){var a=b.createElement("canvas");return!(!a.getContext||!a.getContext("2d"))},C.canvastext=function(){return!(!o.canvas||!f(b.createElement("canvas").getContext("2d").fillText,"function"))},C.webgl=function(){return!!a.WebGLRenderingContext},C.touch=function(){var c;return"ontouchstart"in a||a.DocumentTouch&&b instanceof DocumentTouch?c=!0:H(["@media (",x.join("touch-enabled),("),r,")","{#modernizr{top:9px;position:absolute}}"].join(""),function(a){c=9===a.offsetTop}),c},C.geolocation=function(){return"geolocation"in navigator},C.postmessage=function(){return!!a.postMessage},C.websqldatabase=function(){return!!a.openDatabase},C.indexedDB=function(){return!!j("indexedDB",a)},C.hashchange=function(){return J("hashchange",a)&&(b.documentMode===c||b.documentMode>7)},C.history=function(){return!(!a.history||!history.pushState)},C.draganddrop=function(){var a=b.createElement("div");return"draggable"in a||"ondragstart"in a&&"ondrop"in a},C.websockets=function(){return"WebSocket"in a||"MozWebSocket"in a},C.rgba=function(){return d("background-color:rgba(150,255,150,.5)"),g(t.backgroundColor,"rgba")},C.hsla=function(){return d("background-color:hsla(120,40%,100%,.5)"),g(t.backgroundColor,"rgba")||g(t.backgroundColor,"hsla")},C.multiplebgs=function(){return d("background:url(https://),url(https://),red url(https://)"),/(url\s*\(.*?){3}/.test(t.background)},C.backgroundsize=function(){return j("backgroundSize")},C.borderimage=function(){return j("borderImage")},C.borderradius=function(){return j("borderRadius")},C.boxshadow=function(){return j("boxShadow")},C.textshadow=function(){return""===b.createElement("div").style.textShadow},C.opacity=function(){return e("opacity:.55"),/^0.55$/.test(t.opacity)},C.cssanimations=function(){return j("animationName")},C.csscolumns=function(){return j("columnCount")},C.cssgradients=function(){var a="background-image:",b="gradient(linear,left top,right bottom,from(#9f9),to(white));",c="linear-gradient(left top,#9f9, white);";return d((a+"-webkit- ".split(" ").join(b+a)+x.join(c+a)).slice(0,-a.length)),g(t.backgroundImage,"gradient")},C.cssreflections=function(){return j("boxReflect")},C.csstransforms=function(){return!!j("transform")},C.csstransforms3d=function(){var a=!!j("perspective");return a&&"webkitPerspective"in q.style&&H("@media (transform-3d),(-webkit-transform-3d){#modernizr{left:9px;position:absolute;height:3px;}}",function(b){a=9===b.offsetLeft&&3===b.offsetHeight}),a},C.csstransitions=function(){return j("transition")},C.fontface=function(){var a;return H('@font-face {font-family:"font";src:url("https://")}',function(c,d){var e=b.getElementById("smodernizr"),f=e.sheet||e.styleSheet,g=f?f.cssRules&&f.cssRules[0]?f.cssRules[0].cssText:f.cssText||"":"";a=/src/i.test(g)&&0===g.indexOf(d.split(" ")[0])}),a},C.generatedcontent=function(){var a;return H(["#",r,"{font:0/0 a}#",r,':after{content:"',v,'";visibility:hidden;font:3px/1 a}'].join(""),function(b){a=b.offsetHeight>=3}),a},C.video=function(){var a=b.createElement("video"),c=!1;try{(c=!!a.canPlayType)&&(c=new Boolean(c),c.ogg=a.canPlayType('video/ogg; codecs="theora"').replace(/^no$/,""),c.h264=a.canPlayType('video/mp4; codecs="avc1.42E01E"').replace(/^no$/,""),c.webm=a.canPlayType('video/webm; codecs="vp8, vorbis"').replace(/^no$/,""))}catch(d){}return c},C.audio=function(){var a=b.createElement("audio"),c=!1;try{(c=!!a.canPlayType)&&(c=new Boolean(c),c.ogg=a.canPlayType('audio/ogg; codecs="vorbis"').replace(/^no$/,""),c.mp3=a.canPlayType("audio/mpeg;").replace(/^no$/,""),c.wav=a.canPlayType('audio/wav; codecs="1"').replace(/^no$/,""),c.m4a=(a.canPlayType("audio/x-m4a;")||a.canPlayType("audio/aac;")).replace(/^no$/,""))}catch(d){}return c},C.localstorage=function(){try{return localStorage.setItem(r,r),localStorage.removeItem(r),!0}catch(a){return!1}},C.sessionstorage=function(){try{return sessionStorage.setItem(r,r),sessionStorage.removeItem(r),!0}catch(a){return!1}},C.webworkers=function(){return!!a.Worker},C.applicationcache=function(){return!!a.applicationCache},C.svg=function(){return!!b.createElementNS&&!!b.createElementNS(B.svg,"svg").createSVGRect},C.inlinesvg=function(){var a=b.createElement("div");return a.innerHTML="<svg/>",(a.firstChild&&a.firstChild.namespaceURI)==B.svg},C.smil=function(){return!!b.createElementNS&&/SVGAnimate/.test(w.call(b.createElementNS(B.svg,"animate")))},C.svgclippaths=function(){return!!b.createElementNS&&/SVGClipPath/.test(w.call(b.createElementNS(B.svg,"clipPath")))};for(var L in C)m(C,L)&&(l=L.toLowerCase(),o[l]=C[L](),F.push((o[l]?"":"no-")+l));return o.input||k(),o.addTest=function(a,b){if("object"==typeof a)for(var d in a)m(a,d)&&o.addTest(d,a[d]);else{if(a=a.toLowerCase(),o[a]!==c)return o;b="function"==typeof b?b():b,"undefined"!=typeof p&&p&&(q.className+=" "+(b?"":"no-")+a),o[a]=b}return o},d(""),s=u=null,function(a,b){function c(a,b){var c=a.createElement("p"),d=a.getElementsByTagName("head")[0]||a.documentElement;return c.innerHTML="x<style>"+b+"</style>",d.insertBefore(c.lastChild,d.firstChild)}function d(){var a=s.elements;return"string"==typeof a?a.split(" "):a}function e(a){var b=r[a[p]];return b||(b={},q++,a[p]=q,r[q]=b),b}function f(a,c,d){if(c||(c=b),k)return c.createElement(a);d||(d=e(c));var f;return f=d.cache[a]?d.cache[a].cloneNode():o.test(a)?(d.cache[a]=d.createElem(a)).cloneNode():d.createElem(a),!f.canHaveChildren||n.test(a)||f.tagUrn?f:d.frag.appendChild(f)}function g(a,c){if(a||(a=b),k)return a.createDocumentFragment();c=c||e(a);for(var f=c.frag.cloneNode(),g=0,h=d(),i=h.length;i>g;g++)f.createElement(h[g]);return f}function h(a,b){b.cache||(b.cache={},b.createElem=a.createElement,b.createFrag=a.createDocumentFragment,b.frag=b.createFrag()),a.createElement=function(c){return s.shivMethods?f(c,a,b):b.createElem(c)},a.createDocumentFragment=Function("h,f","return function(){var n=f.cloneNode(),c=n.createElement;h.shivMethods&&("+d().join().replace(/[\w\-]+/g,function(a){return b.createElem(a),b.frag.createElement(a),'c("'+a+'")'})+");return n}")(s,b.frag)}function i(a){a||(a=b);var d=e(a);return!s.shivCSS||j||d.hasCSS||(d.hasCSS=!!c(a,"article,aside,dialog,figcaption,figure,footer,header,hgroup,main,nav,section{display:block}mark{background:#FF0;color:#000}template{display:none}")),k||h(a,d),a}var j,k,l="3.7.0",m=a.html5||{},n=/^<|^(?:button|map|select|textarea|object|iframe|option|optgroup)$/i,o=/^(?:a|b|code|div|fieldset|h1|h2|h3|h4|h5|h6|i|label|li|ol|p|q|span|strong|style|table|tbody|td|th|tr|ul)$/i,p="_html5shiv",q=0,r={};!function(){try{var a=b.createElement("a");a.innerHTML="<xyz></xyz>",j="hidden"in a,k=1==a.childNodes.length||function(){b.createElement("a");var a=b.createDocumentFragment();return"undefined"==typeof a.cloneNode||"undefined"==typeof a.createDocumentFragment||"undefined"==typeof a.createElement}()}catch(c){j=!0,k=!0}}();var s={elements:m.elements||"abbr article aside audio bdi canvas data datalist details dialog figcaption figure footer header hgroup main mark meter nav output progress section summary template time video",version:l,shivCSS:m.shivCSS!==!1,supportsUnknownElements:k,shivMethods:m.shivMethods!==!1,type:"default",shivDocument:i,createElement:f,createDocumentFragment:g};a.html5=s,i(b)}(this,b),o._version=n,o._prefixes=x,o._domPrefixes=A,o._cssomPrefixes=z,o.mq=I,o.hasEvent=J,o.testProp=function(a){return h([a])},o.testAllProps=j,o.testStyles=H,o.prefixed=function(a,b,c){return b?j(a,b,c):j(a,"pfx")},q.className=q.className.replace(/(^|\s)no-js(\s|$)/,"$1$2")+(p?" js "+F.join(" "):""),o}(this,this.document);

      // GA Segmentation
      window._gaq = window._gaq || []; _gaq.push(['_setAccount', 'UA-2330466-1']);_gaq.push(['_setDomainName', '.slideshare.net']);_gaq.push(['_addIgnoredRef', 'slideshare.net']);_gaq.push(['_setCustomVar', 1, 'member_type', 'LOGGEDOUT', 1]);
        _gaq.push(['_trackPageview']);

      // Comscore
      var _comscore = _comscore || [];
      _comscore.push({ c1: "2", c2: "6402952" });

      // Slideshare object
      var slideshare_object = {
        user: {"is_pro":"false","is_valid_fbuser":false,"loggedin":false,"is_test_user":false,"is_li_connected":false,"userGroup":"non-member","id":null,"show_li_connect_cta":false,"li_tracking_url":"https://www.linkedin.com/li/track","login":"guest","fb_userid":null,"su":false,"has_uploads":null,"name":null,"has_privacy_enabled":null},
        timer: { start: (new Date()).getTime(), end: '', execTime: '' },
        top_nav: { get_url: "/top_nav"},
        dev: false,
        init: [],
        feature_flag: [],
        is_mobile: "",
        deploy_environment: "production",
        rum_pagekey: "desktop_slideview_loggedout",
        is_ssl: false
      };
    </script>

    <script id="page-json" type="text/javascript"><!--
 var sso_redirect_uri = "https://www.linkedin.com/uas/oauth2/authorization?bypass_dialog=true\u0026client_id=y4wa9oe4c6nu\u0026redirect_uri=https%3A%2F%2Fwww.slideshare.net%2Fsso%2Fredirect\u0026response_type=code\u0026scope=\u0026state=72bf0a5f40e85ca2ace96f41516057"; 
//--></script>

    <meta content="http://public.slidesharecdn.com/b/images/artdeco/icons.svg?43e81fd2ef" name="ss-svg-icons" />

    <script src="http://public.slidesharecdn.com/b/ss_foundation/combined_experiments.js?0e4ab00b14" type="text/javascript"></script>

    

      <link rel="dns-prefetch" href="//js.bizographics.com">
  <script id="adQueue" type="text/javascript"><!--
 if(!slideshare_object.delayedLIAd){ slideshare_object._adQueue = []; } 
//--></script>
  <script type="text/javascript" id="ga-init">
  // <![CDATA[
    // Setting custom variable [slot = 3, scope = 3(page level)] for tracking source (related, search etc) of pageview
    window._gaq.push(['_setCustomVar', 3, 'source', 'not_set', 3]);
    window._gaq.push(['_trackEvent', 'bigfoot_slideview', 'pageload', slideshare_object.gam_cat_name, undefined, true]);

  // ]]>
    <!--
 slideshare_object.deploy_environment = 'production'; 
//-->
  </script>
    <link href="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" rel="canonical" />
  <link title="Slideshow json oEmbed Profile" type="application/json+oembed" href="http://www.slideshare.net/api/oembed/2?format=json&amp;url=http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" rel="alternate" />
  <link title="Slideshow xml oEmbed Profile" type="text/xml+oembed" href="http://www.slideshare.net/api/oembed/2?format=xml&amp;url=http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" rel="alternate" />
  <link media="handheld" href="http://www.slideshare.net/mobile/ToruUenoyama/tensorflow-gdg" rel="alternate" />

    <link href="android-app://net.slideshare.mobile/slideshare-app/ss/58792278" rel="alternate" />
    <link href="ios-app://917418728/slideshare-app/ss/58792278" rel="alternate" />
  <meta name="apple-itunes-app" content="app-id=917418728, affiliate-data=ct=smart_banner&amp;pt=10746, app-argument=slideshare-app://ss/58792278"/>
  <link rel="manifest" href="/manifest.json"/>


  <!-- fb open graph meta tags -->
    <meta content="2490221586" class="fb_og_meta" property="fb:app_id" name="fb_app_id" />
    <meta content="slideshare:presentation" class="fb_og_meta" property="og:type" name="og_type" />
      <meta content="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" class="fb_og_meta" property="og:url" name="og_url" />
    <meta content="http://cdn.slidesharecdn.com/ss_thumbnails/machinelearningwithtensorflow20160227-160227125739-thumbnail-4.jpg?cb=1456578209" class="fb_og_meta" property="og:image" name="og_image" />
    <!-- SL:start:notranslate -->
    <meta content="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会)" class="fb_og_meta" property="og:title" name="og_title" />
    <meta content="TensorFlow の基本的な使い方やコーディング方法を、線形重回帰やロジスティック回帰、パーセプトロンの実装例を通じて解説しています。（機械学習初心者向け） ※2016/02/27に開催した「GDG京都 機械学習勉強会」で発表した資料です。" class="fb_og_meta" property="og:description" name="og_description" />
    <!-- SL:end:notranslate -->
    <meta content="2016-02-27T12:57:39Z" class="fb_og_meta" property="slideshare:published" name="slideshow_published_time" />
    <meta content="http://www.slideshare.net/ToruUenoyama" class="fb_og_meta" property="slideshare:author" name="slideshow_author" />
    <meta content="327" class="fb_og_meta" property="slideshare:view_count" name="slideshow_view_count" />
    <meta content="0" class="fb_og_meta" property="slideshare:embed_count" name="slideshow_embed_count" />
    <meta content="0" class="fb_og_meta" property="slideshare:comment_count" name="slideshow_comment_count" />
    <meta content="4" class="fb_og_meta" property="slideshare:download_count" name="slideshow_download_count" />
    <meta content="2016-02-27 12:57:39 UTC" class="fb_og_meta" property="slideshare:created_at" name="slideshow_created_at" />
    <meta content="2016-02-27 13:03:29 UTC" class="fb_og_meta" property="slideshare:updated_at" name="slideshow_updated_at" />
    <meta content="" class="fb_og_meta" property="slideshare:featured_on" name="slideshow_featured_on" />
    <meta content="6" class="fb_og_meta" property="slideshare:favorites_count" name="slideshow_favorites_count" />
    <meta content="Technology" class="fb_og_meta" property="slideshare:category" name="slideshow_category" />

  <!-- SL:start:notranslate -->
    <meta name="twitter:card" value="player" />
    <meta name="twitter:site" value="@slideshare" />
    <meta class="twitter_player" value="https://www.slideshare.net/slideshow/embed_code/key/tkPJMfS5uuYI0N" name="twitter:player" />
    <meta name="twitter:player:width" value="342" />
    <meta name="twitter:player:height" value="291" />
    <meta class="twitter_title" value="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会)" name="twitter:title" />
    <meta class="twitter_image" value="https://cdn.slidesharecdn.com/ss_thumbnails/machinelearningwithtensorflow20160227-160227125739-thumbnail-4.jpg?cb=1456578209" name="twitter:image" />
    <meta name="twitter:app:name:googleplay" content="SlideShare Android"/>
    <meta name="twitter:app:id:googleplay" content="net.slideshare.mobile"/>
    <meta name="twitter:app:url:googleplay" content="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" />
    <meta name="twitter:app:name:iphone" content="SlideShare iOS" />
    <meta name="twitter:app:id:iphone" content="917418728" />
    <meta name="twitter:app:url:iphone" content="slideshare-app://ss/58792278" />
    <meta name="twitter:app:name:ipad" content="SlideShare iOS" />
    <meta name="twitter:app:id:ipad" content="917418728" />
    <meta name="twitter:app:url:ipad" content="slideshare-app://ss/58792278" />

  <meta property="al:android:url" content="slideshare-app://ss/58792278" />
  <meta property="al:android:app_name" content="SlideShare Android" />
  <meta property="al:android:package" content="net.slideshare.mobile" />
  <meta property="al:ios:url" content="slideshare-app://ss/58792278" />
  <meta property="al:ios:app_store_id" content="917418728" />
  <meta property="al:ios:app_name" content="SlideShare iOS" />
  <!-- SL:end:notranslate -->

  </head>

  <body id="pagekey-slideshare_desktop_slideview_loggedout" class=" ">



      <div id="main-nav" class="contain-to-grid fixed">
        <div class="icon-bar four-up hide-for-medium-up icon-bar-custom">
          <a class="item" href="/" aria-labelledby="#home">
            <i class="icon-slideshare-logo"></i>
            <label id="home">SlideShare</label>
          </a>
          <a class="item" href="/explore" aria-labelledby="#explore">
            <i class="fa fa-compass"></i>
            <label id="explore">Explore</label>
          </a>
          <a class="item j-open-mobile-search" aria-labelledby="#search">
            <i class="fa fa-search"></i>
            <label id="search">Search</label>
          </a>
            <a class="item" class="void_fancybox void_redirect_link" href="https://www.slideshare.net/login" aria-labelledby="#you">
              <i class="fa fa-user"></i>
              <label id="you">You</label>
            </a>
        </div>


        <nav class="top-bar visible-for-medium-up" data-topbar>
          <ul class="title-area">
            <li class="name">
              <a href="/">
                <img id="main-navbar-logo" alt="LinkedIn SlideShare" src="http://public.slidesharecdn.com/b/images/logo/linkedin-ss/SS_Logo_White_Large.png?6d1f7a78a6">
              </a>
            </li>
          </ul>

          <section class="top-bar-section">
            <ul class="right upload-create button-group">
                <li class="has-form upload-create">
                  <a href="/upload" class="button radius secondary">Upload</a>
                </li>

                <li>
                  <a id="login" class="void_fancybox void_redirect_link" href="https://www.slideshare.net/login">Login</a>
                </li>
                <li>
                  <a id="signup" href="https://www.slideshare.net/signup" class="void_fancybox void_redirect_link" title="Signup now for a SlideShare account">Signup</a>
                </li>
            </ul>

            <ul class="left">
              <li class="divider"></li>
              <li id="nav-search" class="has-form">
                <form id="nav-search-form" method="get" action="/search/slideshow">
                  <div class="input-box">
                  <input name="searchfrom" type="hidden" value="header">
                    <input id="nav-search-query" type="text" placeholder="Search" name="q" value="" autocomplete="off">
                    <button class="button expand" type="submit"><i class="fa fa-search"></i></button>
                    <ul class='search-suggestions dropdown f-dropdown medium' style="display:none;"></ul>
                  </div>
                </form>
              </li>
            </ul>
          </section>
        </nav>

          <div class="visible-for-medium-up sub-navbar">
            <div class="row">
              <div class="container">
                <ul class="sub-nav-cats left">
                    <li class="sub-nav-cat">
                      <a href="/" class="sub-nav-link" data-ga-action="click" data-ga-cat="sub_nav_cat" data-ga-label="Home" rel="nofollow">Home</a>
                    </li>
                    <li class="sub-nav-cat">
                      <a href="/featured/category/technology" class="sub-nav-link" data-ga-action="click" data-ga-cat="sub_nav_cat" data-ga-label="Technology" rel="nofollow">Technology</a>
                    </li>
                    <li class="sub-nav-cat">
                      <a href="/featured/category/education" class="sub-nav-link" data-ga-action="click" data-ga-cat="sub_nav_cat" data-ga-label="Education" rel="nofollow">Education</a>
                    </li>
                    <li class="sub-nav-cat">
                      <a href="/explore" class="sub-nav-link" data-ga-action="click" data-ga-cat="sub_nav_cat" data-ga-label="More Topics" rel="nofollow">More Topics</a>
                    </li>
                </ul>
                <ul class="sub-nav-cats right show-for-large-up sub-nav-link-ctas">
                  <li class="sub-nav-cat has-dropdown creators-hub-dropdown" data-dropdown="creators-hub-dropdown" data-options="is_hover:true;">
                    <a href="/ss/creators?from=sub-nav" class="sub-nav-link creators-hub" data-ga-action="click" data-ga-cat="sub_nav_cat" data-ga-label="creators_hub">For Uploaders</a>
                  </li>
                  <li class="sub-nav-cat collect-leads-cta">
                  <a href="/lead-generation" class="sub-nav-link" data-ga-action="click" data-ga-cat="sub_nav_cat" data-ga-label="Collect Leads">Collect Leads</a>
                  </li>
                </ul>
                <ul id="creators-hub-dropdown" class="dropdown f-dropdown creators-hub-dropdown" data-dropdown-content>
                  <li>
                    <a href="/ss/creators/get-started?from=sub-nav"
                    data-ga-cat="sub_nav_cat"
                    data-ga-action="click"
                    data-ga-label="creators_hub_gs"
                    rel="nofollow">
                    Get Started</a>
                  </li>
                  <li class="divider"></li>
                  <li>
                    <a href="/ss/creators/tips-and-tricks?from=sub-nav"
                    data-ga-cat="sub_nav_cat"
                    data-ga-action="click"
                    data-ga-label="creators_hub_tat"
                    rel="nofollow">
                    Tips & Tricks</a>
                  </li>
                  <li class="divider"></li>
                  <li>
                    <a href="/ss/creators/tools?from=sub-nav"
                    data-ga-cat="sub_nav_cat"
                    data-ga-action="click"
                    data-ga-label="creators_hub_tools"
                    rel="nofollow">
                    Tools</a>
                  </li>
                  <li class="divider"></li>
                  <li>
                    <a href="/ss/creators/for-businesses?from=sub-nav"
                    data-ga-cat="sub_nav_cat"
                    data-ga-action="click"
                    data-ga-label="creators_hub_bus"
                    rel="nofollow">
                    For Business</a>
                  </li>
                </ul>
              </div>
            </div>
          </div>
      </div>

    <div class="wrapper">
        <div id="main-nav-mobile-search">
          <form name="mobile-search" action="/search/slideshow" method="get">
           <div class="input-box">
              <input type="text" placeholder="Search" name="q" value="" autocomplete="off">
              <a class="button expand">
                <i class="search-icon fa fa-search"></i>
                <i class="search-spinner fa fa-spinner fa-spin"></i>
              </a>
            </div>
          </form>
        </div>
        


      

      




<iframe id="download" style="display:none" src="" class="notify"></iframe>
<div id="j-downloading-alert" data-alert class="alert-box success radius
downloading-alert">
  <span class="copy-in-aria-label" aria-label="Your SlideShare is downloading."></span>
  <a href="#" class="close copy-in-aria-label" aria-label="&times;"></a>
</div>





<div id="slideview-container" class="">
  <div class="slideview-header-container row">
    <div class="columns">

        <div id="top-ad" class="j-top-ad banner">
          <div id="leaderboardAd" class="leaderboard1" style="max-height:15px">
          </div>
        </div>



      <div></div>
    </div>
  </div>

  <div class="row">

    <div id="main-panel" class="small-12 large-8 columns">


      <div class="sectionElements">

          

<div class="j-clips-toolbar clips-toolbar ">

  <div class="clips-progress progress">
    <span class="j-clips-meter clip-meter meter" style="width:0%"></span>
  </div>

  <div class="j-clips-info clips-info">
    <span class="j-clips-count clips-count copy-in-aria-label" aria-label="0"></span>
    <span class="j-clips-count-text clips-count-text notranslate copy-in-aria-label"></span>
  </div>

</div>


        <div class="playerWrapper">



            <div>
              <!-- For slideview page , combined js for player is now combined with slideview javascripts for logged out users-->


<div class="player lightPlayer fluidImage presentation_player" id="svPlayerId">
  <span class="j-fullscreen-title-banner fullscreen-title-banner" style="display: none;">
    TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会)
  </span>

  <div class="stage valign-first-slide">
    <div class="player-cta-container">
          <div class="clip-button-top">
  <a id="clips-button-top" class="j-clips-button clip-button player-cta visible "
    href="/signup?login_source=slideview.clip.like&amp;from=clip&amp;layout=foundation&amp;from_source="
    rel="nofollow" data-reveal-id=login_modal style="display:none">
    <div class="container clearfix">
      <div class="svg-icon">
        <svg><use data-size="small" xlink:href="#clipboard-add-icon"></use></svg>
      </div>
      <span class="clip-button-text-clip notranslate copy-in-aria-label" aria-label="Clip slide" title="Clip to save this slide for later"></span>
    </div>
  </a>
</div>

    </div>
    <a class="exit-fullscreen j-exit-fullscreen" style="display: none;">
      <i class="fa fa-compress fa-2x"></i>
    </a>
    <div class="slide_container">
            <section data-index="1"  class="slide show" itemprop=image>
                <img class="slide_image"
                  src="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-1-638.jpg?cb=1456578209"
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-1-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-1-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-1-1024.jpg?cb=1456578209"
                  alt="TensorFlow を使った
機械学習ことはじめ
2016-02-27
GDG京都 上野山
 "/>
            </section>
            <section data-index="2"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-2-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-2-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-2-1024.jpg?cb=1456578209"
                  alt="自己紹介
 電機メーカ勤務の
似非ソフト屋
 もともと組込み
 画像処理
 データ解析 と変遷
 最近は Android より
R言語(統計処理言語)と
格闘中
2
上野山 徹
うえのやま とおる
 "/>
            </section>
            <section data-index="3"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-3-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-3-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-3-1024.jpg?cb=1456578209"
                  alt="1. TensorFlow とは
2016-02-27 GDG京都 機械学習勉強会
3
 "/>
            </section>
            <section data-index="4"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-4-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-4-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-4-1024.jpg?cb=1456578209"
                  alt="TensorFlow とは
 Google社のMachine Intelligence
研究所が開発した数値計算ライブラリ
 Deep Learning をはじめとする 機械学習 の
研究や活用にGoogle社内で用いられている
 201..."/>
            </section>
            <section data-index="5"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-5-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-5-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-5-1024.jpg?cb=1456578209"
                  alt="データフローグラフ
2016-02-27 GDG京都 機械学習勉強会
5
 掛け算
 足し算
 関数
テンソル が データフローグラフ 上を
流れるように計算処理している
テンソル 
(多次元配列)
𝐶 = … (𝑅𝑒𝐿𝑈 𝑏 + 𝑊𝑋..."/>
            </section>
            <section data-index="6"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-6-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-6-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-6-1024.jpg?cb=1456578209"
                  alt="TensorBoard
 データフローグラフのネットワーク構造や
学習経過をWeb上で確認できる
2016-02-27 GDG京都 機械学習勉強会
6
 "/>
            </section>
            <section data-index="7"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-7-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-7-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-7-1024.jpg?cb=1456578209"
                  alt="詳細は Web へ
2016-02-27 GDG京都 機械学習勉強会
7
https://speakerdeck.com/kazunori279/cloud-vision-api-and-tensorflow
例えば、佐藤さん@Google の..."/>
            </section>
            <section data-index="8"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-8-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-8-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-8-1024.jpg?cb=1456578209"
                  alt="2. TensorFlow を使った
機械学習
2016-02-27 GDG京都 機械学習勉強会
8
 "/>
            </section>
            <section data-index="9"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-9-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-9-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-9-1024.jpg?cb=1456578209"
                  alt="さっそく問題です
2016-02-27 GDG京都 機械学習勉強会
9
たかしくんは八百屋へお使いに行きました。
リンゴ１個とミカン３個を買うと１９０円，
リンゴ３個とミカン１個を買うと３３０円
するようです。
リンゴ２個とミカン４個を買うと
..."/>
            </section>
            <section data-index="10"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-10-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-10-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-10-1024.jpg?cb=1456578209"
                  alt="円
式で表すと...
2016-02-27 GDG京都 機械学習勉強会
10
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
 "/>
            </section>
            <section data-index="11"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-11-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-11-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-11-1024.jpg?cb=1456578209"
                  alt="円
回答
2016-02-27 GDG京都 機械学習勉強会
11
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
100 30
100 30
100 30
 "/>
            </section>
            <section data-index="12"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-12-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-12-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-12-1024.jpg?cb=1456578209"
                  alt="円
回答
2016-02-27 GDG京都 機械学習勉強会
12
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ 320
100 30
100 30
100 30
 "/>
            </section>
            <section data-index="13"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-13-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-13-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-13-1024.jpg?cb=1456578209"
                  alt="円
これも機械学習（回帰）の問題です！
2016-02-27 GDG京都 機械学習勉強会
13
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ 320
100 30
100..."/>
            </section>
            <section data-index="14"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-14-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-14-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-14-1024.jpg?cb=1456578209"
                  alt="実は店のおやじは気まぐれでした
2016-02-27 GDG京都 機械学習勉強会
14
円
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
円 × 5 ＋ 円 × 7 ..."/>
            </section>
            <section data-index="15"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-15-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-15-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-15-1024.jpg?cb=1456578209"
                  alt="実は店のおやじは気まぐれでした
2016-02-27 GDG京都 機械学習勉強会
15
円
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
円 × 5 ＋ 円 × 7 ..."/>
            </section>
            <section data-index="16"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-16-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-16-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-16-1024.jpg?cb=1456578209"
                  alt="どうやってパラメータ(単価)を見つけるか？
2016-02-27 GDG京都 機械学習勉強会
16
考え方
合計金額の誤差が小さくなるような
単価を採用しよう!
円 × 1 ＋ 円 × 3  40円 - 190 円
円 × 3 ＋ 円 × 1..."/>
            </section>
            <section data-index="17"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-17-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-17-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-17-1024.jpg?cb=1456578209"
                  alt="すべての組合せを計算
2016-02-27 GDG京都 機械学習勉強会
17
ミカン
0 10 20 30 40 50 60 70 80 90 100
リンゴ
0 193533 158700 127800 100833 77800 58700 ..."/>
            </section>
            <section data-index="18"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-18-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-18-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-18-1024.jpg?cb=1456578209"
                  alt="よく見ると、徐々に変化している
2016-02-27 GDG京都 機械学習勉強会
18
ミカン
0 10 20 30 40 50 60 70 80 90 100
リンゴ
0 193533 158700 127800 100833 77800 5..."/>
            </section>
            <section data-index="19"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-19-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-19-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-19-1024.jpg?cb=1456578209"
                  alt="小さくなる方向へたどっていくと...
2016-02-27 GDG京都 機械学習勉強会
19
ミカン
0 10 20 30 40 50 60 70 80 90 100
リンゴ
0 193533
10 132733
20 83667
30 463..."/>
            </section>
            <section data-index="20"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-20-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-20-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-20-1024.jpg?cb=1456578209"
                  alt="気まぐれおやじのこころを読む
2016-02-27 GDG京都 機械学習勉強会
20
円
円 × 1 ＋ 円 × 3 ≒ 190 円
円 × 3 ＋ 円 × 1 ≒ 330 円
円 × 2 ＋ 円 × 4 ≒
円 × 5 ＋ 円 × 7 ≒ 6..."/>
            </section>
            <section data-index="21"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-21-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-21-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-21-1024.jpg?cb=1456578209"
                  alt="ここまでの振り返り
2016-02-27 GDG京都 機械学習勉強会
21
TensorFlowでも同じように書いていく
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタ..."/>
            </section>
            <section data-index="22"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-22-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-22-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-22-1024.jpg?cb=1456578209"
                  alt="線形回帰(線形重回帰)の1つ
2016-02-27 GDG京都 機械学習勉強会
22
入力層 出力層
𝑦
𝑥1
𝑥2
× 𝑎1
𝑦 = 𝑎1 𝑥1 + 𝑎2 𝑥2
× 𝑎2
リンゴの数 
ミカンの数 
合計金額
リンゴ
の単価
↓
↑
ミ..."/>
            </section>
            <section data-index="23"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-23-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-23-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-23-1024.jpg?cb=1456578209"
                  alt="TensorFlow の処理の流れ
2016-02-27 GDG京都 機械学習勉強会
23
# coding: utf-8
# 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplot ..."/>
            </section>
            <section data-index="24"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-24-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-24-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-24-1024.jpg?cb=1456578209"
                  alt="0. ライブラリをロード
2016-02-27 GDG京都 機械学習勉強会
24
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
Tensor..."/>
            </section>
            <section data-index="25"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-25-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-25-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-25-1024.jpg?cb=1456578209"
                  alt="1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
25
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=(None, 2), name=&quot;x&quot;..."/>
            </section>
            <section data-index="26"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-26-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-26-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-26-1024.jpg?cb=1456578209"
                  alt="2.誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
26
# 誤差関数(loss)
loss = tf.reduce_mean(tf.square(y_ - y))
# 最適化手段を選ぶ(最急降下法)
train_..."/>
            </section>
            <section data-index="27"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-27-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-27-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-27-1024.jpg?cb=1456578209"
                  alt="3. 訓練データを作成(or読込)し，
2016-02-27 GDG京都 機械学習勉強会
27
train_x = np.array([[1., 3.], [3., 1.], [5., 7.]])
train_y = np.array([190..."/>
            </section>
            <section data-index="28"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-28-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-28-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-28-1024.jpg?cb=1456578209"
                  alt="学習実行
2016-02-27 GDG京都 機械学習勉強会
28
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(..."/>
            </section>
            <section data-index="29"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-29-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-29-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-29-1024.jpg?cb=1456578209"
                  alt="参考: 実行結果
2016-02-27 GDG京都 機械学習勉強会
29
うまく学習できると...
step= 10, a1= 70.35, a2= 46.23, loss=2189.06
step= 20, a1= 83.06, a2= 36..."/>
            </section>
            <section data-index="30"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-30-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-30-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-30-1024.jpg?cb=1456578209"
                  alt="参考: 実行結果
2016-02-27 GDG京都 機械学習勉強会
30
学習率が大きすぎて学習失敗していると...
step= 10, (中略), loss=72692296.00
step= 20, (中略), loss=546510766..."/>
            </section>
            <section data-index="31"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-31-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-31-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-31-1024.jpg?cb=1456578209"
                  alt="4. 予測
2016-02-27 GDG京都 機械学習勉強会
31
# (1) 新しいデータを用意
new_x = np.array([2., 4.]).reshape(1, 2)
# (2) 学習結果をつかって，予測実施
new_y = se..."/>
            </section>
            <section data-index="32"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-32-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-32-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-32-1024.jpg?cb=1456578209"
                  alt="5. 後片付け
2016-02-27 GDG京都 機械学習勉強会
32
# セッションを閉じる
sess.close()
 "/>
            </section>
            <section data-index="33"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-33-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-33-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-33-1024.jpg?cb=1456578209"
                  alt="TensorFlow の処理の流れ (再掲)
2016-02-27 GDG京都 機械学習勉強会
33
# coding: utf-8
# 必要なモジュールを読み込む
import numpy as np
import matplotlib.py..."/>
            </section>
            <section data-index="34"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-34-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-34-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-34-1024.jpg?cb=1456578209"
                  alt="(補足) 線形回帰の応用場面
 家賃
 家賃 = a * 駅からの距離 + b * 築年数 + c * 広さ
 明日の客数
 客数 = a * 気温 + b * 昨年同日の客数
などなど
2016-02-27 GDG京都 機械学習勉強会..."/>
            </section>
            <section data-index="35"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-35-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-35-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-35-1024.jpg?cb=1456578209"
                  alt="つぎの問題です
2016-02-27 GDG京都 機械学習勉強会
35
たかしくんは八百屋へ財布を預かってお使いに
行きました。しかしたかしくんはお金を
数えられません。
気まぐれおやじ曰く、
リンゴ２個＋ミカン３個、リンゴ０個＋ミカン１６個
..."/>
            </section>
            <section data-index="36"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-36-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-36-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-36-1024.jpg?cb=1456578209"
                  alt="機械学習の流れ
2016-02-27 GDG京都 機械学習勉強会
36
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 "/>
            </section>
            <section data-index="37"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-37-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-37-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-37-1024.jpg?cb=1456578209"
                  alt="式で表そうとしてみる...
2016-02-27 GDG京都 機械学習勉強会
37
円 ×1＋ 円
円 ×2＋ 円 × 3)
円 ×0＋ 円 ×16)
円 ×3＋ 円 × 1)
円
円
円
円 ×2＋ 円 × 8 )円
×11)円
-(
-(
..."/>
            </section>
            <section data-index="38"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-38-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-38-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-38-1024.jpg?cb=1456578209"
                  alt="式で表そうとしてみる...
2016-02-27 GDG京都 機械学習勉強会
38
円 ×1＋ 円
円 ×2＋ 円 × 3)
円 ×0＋ 円 ×16)
円 ×3＋ 円 × 1)
円
円
円
円 ×2＋ 円 × 8 )円
×11)円
-(
-(
..."/>
            </section>
            <section data-index="39"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-39-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-39-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-39-1024.jpg?cb=1456578209"
                  alt="シグモイド曲線
2016-02-27 GDG京都 機械学習勉強会
39
予想される残金
0
0.5
買える
買えない
𝜎 𝑢 =
1
1 + exp(−𝑢)
ロジット
 "/>
            </section>
            <section data-index="40"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-40-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-40-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-40-1024.jpg?cb=1456578209"
                  alt="予測式(モデル)が作れた！
2016-02-27 GDG京都 機械学習勉強会
40
円 ×1＋ 円
円 ×2＋ 円 × 3)) ＝ 1
円 ×0＋ 円 ×16)) ＝ 1
円 ×3＋ 円 × 1)) ＝ 0
円
円
円
円 ×2＋ 円 × 8 ..."/>
            </section>
            <section data-index="41"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-41-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-41-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-41-1024.jpg?cb=1456578209"
                  alt="ロジスティック回帰
2016-02-27 GDG京都 機械学習勉強会
41
入力層 出力層
※ロジスティック「回帰」ですが，識別(分類)の手法です。
𝑦 = 𝜎(𝑏 + 𝑎1 𝑥1 + 𝑎2 𝑥2)
𝑦
𝑥1
𝑥2
× 𝑎1
× 𝑎2
リンゴの..."/>
            </section>
            <section data-index="42"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-42-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-42-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-42-1024.jpg?cb=1456578209"
                  alt="TensorFlow での実装
2016-02-27 GDG京都 機械学習勉強会
42
# coding: utf-8
# ### 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplo..."/>
            </section>
            <section data-index="43"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-43-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-43-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-43-1024.jpg?cb=1456578209"
                  alt="1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
43
# 1. 学習したいモデルを記述する
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=..."/>
            </section>
            <section data-index="44"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-44-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-44-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-44-1024.jpg?cb=1456578209"
                  alt="2. 誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
44
# 誤差関数(loss)
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(u,..."/>
            </section>
            <section data-index="45"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-45-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-45-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-45-1024.jpg?cb=1456578209"
                  alt="3. 学習実行
2016-02-27 GDG京都 機械学習勉強会
45
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.r..."/>
            </section>
            <section data-index="46"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-46-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-46-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-46-1024.jpg?cb=1456578209"
                  alt="学習結果
2016-02-27 GDG京都 機械学習勉強会
46
step=100, a1=-22.50, a2=-12.28, b=196.26, loss=42.75
step=200, a1=-35.00, a2=-12.06, b=19..."/>
            </section>
            <section data-index="47"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-47-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-47-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-47-1024.jpg?cb=1456578209"
                  alt="予測結果
2016-02-27 GDG京都 機械学習勉強会
47
円 ×1＋ 円 ×11) ) ＝円 -( 0.96σ( 185 60 11
リンゴ１個＋ミカン１１個は，おそらく買えそう
 "/>
            </section>
            <section data-index="48"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-48-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-48-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-48-1024.jpg?cb=1456578209"
                  alt="最後にもう少し複雑な問題
2016-02-27 GDG京都 機械学習勉強会
48
手書き文字認識
MNISTデータ
0～9の手書き文字画像を7万枚集めたデータセット
1枚の画像は 28 x 28 ピクセル
7 3 4 6
多クラス識別(分類)問..."/>
            </section>
            <section data-index="49"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-49-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-49-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-49-1024.jpg?cb=1456578209"
                  alt="機械学習の流れ
2016-02-27 GDG京都 機械学習勉強会
49
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 "/>
            </section>
            <section data-index="50"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-50-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-50-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-50-1024.jpg?cb=1456578209"
                  alt="one-hot ベクトル (one-of-K表現)
2016-02-27 GDG京都 機械学習勉強会
50
7
3
4
6
出力ラベル(1次元) 出力ラベル(10次元)
０ 0 0 0 0 0 0 1 0 0
0 1 2 3 4 5 6 7 8..."/>
            </section>
            <section data-index="51"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-51-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-51-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-51-1024.jpg?cb=1456578209"
                  alt="案1: ロジスティック回帰を拡張
2016-02-27 GDG京都 機械学習勉強会
51
入力層
(768ノード)
出力層
(10ノード)
𝑦0
𝑦1
𝑦9
ソフトマックス(Softmax) 𝑦 = 𝑠𝑜𝑓𝑡𝑚𝑎𝑥(𝑥𝑊 + 𝑏)
 "/>
            </section>
            <section data-index="52"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-52-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-52-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-52-1024.jpg?cb=1456578209"
                  alt="案1: ロジスティック回帰を拡張
2016-02-27 GDG京都 機械学習勉強会
52
入力層
(768ノード)
出力層
(10ノード)
𝑦0
𝑦1
𝑦9
ソフトマックス(Softmax)
 "/>
            </section>
            <section data-index="53"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-53-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-53-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-53-1024.jpg?cb=1456578209"
                  alt="案2: さらに中間層を追加
2016-02-27 GDG京都 機械学習勉強会
53
入力層
(768ノード)
出力層
(10ノード)
𝑦0
𝑦1
𝑦9
ニューラルネット（パーセプトロン）
隠れ層
(□ノード)
中間層(隠れ層)を追加すると，より..."/>
            </section>
            <section data-index="54"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-54-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-54-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-54-1024.jpg?cb=1456578209"
                  alt="TensorFlow での実装
2016-02-27 GDG京都 機械学習勉強会
54
# 0. 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplot as plt
import t..."/>
            </section>
            <section data-index="55"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-55-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-55-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-55-1024.jpg?cb=1456578209"
                  alt="1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
55
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.plac..."/>
            </section>
            <section data-index="56"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-56-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-56-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-56-1024.jpg?cb=1456578209"
                  alt="2. 誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
56
# 誤差関数(loss)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(u,..."/>
            </section>
            <section data-index="57"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-57-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-57-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-57-1024.jpg?cb=1456578209"
                  alt="3. 学習実行
2016-02-27 GDG京都 機械学習勉強会
57
# MNISTデータの取得
mnist = input_data.read_data_sets(&quot;MNIST_data/&quot;, one_hot=True)
# (2) バッチ..."/>
            </section>
            <section data-index="58"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-58-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-58-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-58-1024.jpg?cb=1456578209"
                  alt="参考: 学習結果
2016-02-27 GDG京都 機械学習勉強会
58
step=1000, loss=2.09
step=2000, loss=1.45
step=3000, loss=1.38
step=4000, loss=0.77
s..."/>
            </section>
            <section data-index="59"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-59-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-59-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-59-1024.jpg?cb=1456578209"
                  alt="4. 予測
2016-02-27 GDG京都 機械学習勉強会
59
# (1) テスト用データを1000サンプル取得
new_x = mnist.test.images[0:1000]
new_y_ = mnist.test.labels[0:..."/>
            </section>
            <section data-index="60"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-60-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-60-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-60-1024.jpg?cb=1456578209"
                  alt="まとめ（機械学習の流れ）
2016-02-27 GDG京都 機械学習勉強会
60
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 "/>
            </section>
            <section data-index="61"  class="slide" >
                <i class="fa fa-spinner fa-spin"></i>
                <img class="slide_image"
                  src=""
                  data-small="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-61-320.jpg?cb=1456578209"
                  data-normal="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-61-638.jpg?cb=1456578209"
                  data-full="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-61-1024.jpg?cb=1456578209"
                  alt="参考書籍
2016-02-27 GDG京都 機械学習勉強会
61
ディープラーニングの基礎的な理論と
TensorFlowなどのサンプルコードあり
ディープラーニングの
理論理解を深めたいかたはこちら。
 "/>
            </section>
        <div class="j-next-container next-container">
          <div class="content-container">
            <div class="next-slideshow-wrapper">
              <div class="j-next-slideshow next-slideshow">
                <div class="title-container">
                  <span class="title-text">Upcoming SlideShare</span>
                </div>
                <div class="j-next-url info">
                  <div class="thumb-container">
                    <img class="j-next-thumb thumb" />
                  </div>
                  <div class="text-container">
                    <span class="j-next-title next-title"></span>
                    <span class="j-next-views next-views"></span>
                  </div>
                </div>
              </div>
              <div class="next-timer">Loading in …<span class="j-timer-count timer-count">5</span></div>
              <div class="j-next-cancel next-cancel">&#215;</div>
            </div>
          </div>
        </div>
    </div>
  </div> <!-- end stage -->

  <div class="toolbar_wrapper j-player-toolbar">
    <div class="toolbar normal">
      
<!-- using div.bar-[top, bottom]-margin to fix toolbar spacing with a taller progressbar (improve slide scrubbing UX) -->
<div class="j-progress-bar progress-bar-wrapper">
  <div class="progress-bar-spacing"></div>
  <div class="buffered-bar"></div>
  <div class="j-slides-loaded-bar progress-bar"></div>
  <div class="j-progress-tooltip progress-tooltip" style="display: none;">
    <div class="j-tooltip-content progress-tooltip-wrapper">
      <img class="j-tooltip-thumb tooltip-thumb" onerror="this.src=''"
            slide-thumb-1=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-1-320.jpg?cb=1456578209
            slide-thumb-2=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-2-320.jpg?cb=1456578209
            slide-thumb-3=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-3-320.jpg?cb=1456578209
            slide-thumb-4=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-4-320.jpg?cb=1456578209
            slide-thumb-5=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-5-320.jpg?cb=1456578209
            slide-thumb-6=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-6-320.jpg?cb=1456578209
            slide-thumb-7=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-7-320.jpg?cb=1456578209
            slide-thumb-8=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-8-320.jpg?cb=1456578209
            slide-thumb-9=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-9-320.jpg?cb=1456578209
            slide-thumb-10=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-10-320.jpg?cb=1456578209
            slide-thumb-11=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-11-320.jpg?cb=1456578209
            slide-thumb-12=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-12-320.jpg?cb=1456578209
            slide-thumb-13=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-13-320.jpg?cb=1456578209
            slide-thumb-14=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-14-320.jpg?cb=1456578209
            slide-thumb-15=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-15-320.jpg?cb=1456578209
            slide-thumb-16=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-16-320.jpg?cb=1456578209
            slide-thumb-17=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-17-320.jpg?cb=1456578209
            slide-thumb-18=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-18-320.jpg?cb=1456578209
            slide-thumb-19=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-19-320.jpg?cb=1456578209
            slide-thumb-20=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-20-320.jpg?cb=1456578209
            slide-thumb-21=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-21-320.jpg?cb=1456578209
            slide-thumb-22=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-22-320.jpg?cb=1456578209
            slide-thumb-23=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-23-320.jpg?cb=1456578209
            slide-thumb-24=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-24-320.jpg?cb=1456578209
            slide-thumb-25=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-25-320.jpg?cb=1456578209
            slide-thumb-26=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-26-320.jpg?cb=1456578209
            slide-thumb-27=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-27-320.jpg?cb=1456578209
            slide-thumb-28=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-28-320.jpg?cb=1456578209
            slide-thumb-29=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-29-320.jpg?cb=1456578209
            slide-thumb-30=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-30-320.jpg?cb=1456578209
            slide-thumb-31=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-31-320.jpg?cb=1456578209
            slide-thumb-32=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-32-320.jpg?cb=1456578209
            slide-thumb-33=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-33-320.jpg?cb=1456578209
            slide-thumb-34=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-34-320.jpg?cb=1456578209
            slide-thumb-35=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-35-320.jpg?cb=1456578209
            slide-thumb-36=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-36-320.jpg?cb=1456578209
            slide-thumb-37=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-37-320.jpg?cb=1456578209
            slide-thumb-38=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-38-320.jpg?cb=1456578209
            slide-thumb-39=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-39-320.jpg?cb=1456578209
            slide-thumb-40=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-40-320.jpg?cb=1456578209
            slide-thumb-41=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-41-320.jpg?cb=1456578209
            slide-thumb-42=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-42-320.jpg?cb=1456578209
            slide-thumb-43=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-43-320.jpg?cb=1456578209
            slide-thumb-44=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-44-320.jpg?cb=1456578209
            slide-thumb-45=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-45-320.jpg?cb=1456578209
            slide-thumb-46=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-46-320.jpg?cb=1456578209
            slide-thumb-47=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-47-320.jpg?cb=1456578209
            slide-thumb-48=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-48-320.jpg?cb=1456578209
            slide-thumb-49=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-49-320.jpg?cb=1456578209
            slide-thumb-50=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-50-320.jpg?cb=1456578209
            slide-thumb-51=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-51-320.jpg?cb=1456578209
            slide-thumb-52=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-52-320.jpg?cb=1456578209
            slide-thumb-53=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-53-320.jpg?cb=1456578209
            slide-thumb-54=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-54-320.jpg?cb=1456578209
            slide-thumb-55=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-55-320.jpg?cb=1456578209
            slide-thumb-56=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-56-320.jpg?cb=1456578209
            slide-thumb-57=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-57-320.jpg?cb=1456578209
            slide-thumb-58=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-58-320.jpg?cb=1456578209
            slide-thumb-59=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-59-320.jpg?cb=1456578209
            slide-thumb-60=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-60-320.jpg?cb=1456578209
            slide-thumb-61=http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/85/tensorflow-gdg-61-320.jpg?cb=1456578209
      >
      <span class="j-slidecount-label slidecount-label">1</span>
    </div>
    <div class="progress-tooltip-caret"></div>
  </div>
</div>
<div class="progress-bar-spacing"></div>



<div class="j-tools bot-actions">
</div><!-- .bot-actions -->


  <div class="j-tools bot-actions">
    <a data-tooltip
      aria-haspopup="true"
      style="display: none"
      class="j-tooltip j-download action-download has-tip"
      title="Save this presentation"
      href="/login?from_source=%2FToruUenoyama%2Ftensorflow-gdg%3Ffrom_action%3Dsave&amp;from=download&amp;layout=foundation"
      data-target="#login_modal"
      data-placement="top">
      <i class="fa fa-download fa-lg" style="margin-top: 1px;"></i>
    </a>
  </div>




  <div class="nav">
      <button id="btnPrevious" title="Previous Slide">
        <div class="j-prev-btn arrow-left disabled"></div>
      </button>
    <label class="goToSlideLabel">
      <span id="current-slide" class="j-current-slide">1</span> of <span id="total-slides" class="j-total-slides">61</span>
    </label>
      <button id="btnNext" title="Next Slide">
        <div class="j-next-btn arrow-right"></div>
      </button>
  </div>



<div class="navActions">



    <button id="btnFullScreen" class="j-tooltip btnFullScreen" title="View Fullscreen">
      <span class="fa fa-stack">
        <i class="fa fa-expand fa-stack-1x"></i>
      </span>
    </button>
    <button id="btnLeaveFullScreen" class="j-tooltip btnLeaveFullScreen" title="Exit Fullscreen">
      <span class="fa-stack">
        <i class="fa fa-compress fa-stack-1x"></i>
      </span>
    </button>

</div>


    </div>
  </div>

  <div class="success-toast-wrapper hide"></div>

  <div class="image_maps"></div>
</div>

<div id="j-lead-form-placeholder" style="display:none">
</div>

            </div>
        </div>

        <div id="lastScreen" style="display: none;">
          <div class="lastScreen">
  <div class="jsplLastScreenOverlay j-last-screen-overlay"></div>
  <div class="pro-overlay j-lastscreen">
    <div class="proSharingText">Like this presentation? Why not share!</div>
    <ul class="lastActions j-last-screen-actions">
      <li class="share-cta j-share-cta lastscreen-share-cta"><a class="share-btn"><span class="lastScreen-sprite"></span>Share</a></li>
      <li class="email-cta j-email-cta"><a class="email-btn"><span class="lastScreen-sprite"></span>Email</a></li>
      <li class="replay last">
        <a class="replay-btn lastScreenReplay j-tooltip j-last-screen-replay" data-original-title="View again" title="View again">
          <span class="lastScreen-sprite">&nbsp;</span>
        </a>
      </li>
      <li class="close-btn lastScreen-sprite j-lastscreen-close">
        <a>&nbsp;</a>
      </li>
    </ul>
    <div class="related-presentations j-lastscreen-related">
      <ul class="presentation-list">
        <li>
         <a href="/mikecrabb/introduction-to-information-architecture-58750805" title="Introduction to Information Archite..." >
           <img class="j-thumbnail" data-original="//cdn.slidesharecdn.com/ss_thumbnails/introductiontoinformationarchitecture-160226103935-thumbnail.jpg?cb=1456521962" alt="Introduction to Information Archite..." />
           <span class="presentation-meta">
             <span class="title">Introduction to Information Archite...</span>
             <span class="author">by&nbsp;Mike Crabb</span>
             <span class="view-count">5710&nbsp;views</span>
           </span>
         </a>
        </li>
        <li>
         <a href="/omerhameed/the-link-between-alcohol-and-breast-cancer" title="The Link Between Alcohol and Breast..." >
           <img class="j-thumbnail" data-original="//cdn.slidesharecdn.com/ss_thumbnails/risk-of-breast-cancer-160226020800-thumbnail.jpg?cb=1456521687" alt="The Link Between Alcohol and Breast..." />
           <span class="presentation-meta">
             <span class="title">The Link Between Alcohol and Breast...</span>
             <span class="author">by&nbsp;Dr.  Omer Hameed</span>
             <span class="view-count">2577&nbsp;views</span>
           </span>
         </a>
        </li>
        <li>
         <a href="/IpsosMORI/the-death-of-polling" title="The Death of Polling?" >
           <img class="j-thumbnail" data-original="//cdn.slidesharecdn.com/ss_thumbnails/thedeathofpolling-160226121628-thumbnail.jpg?cb=1456521635" alt="The Death of Polling?" />
           <span class="presentation-meta">
             <span class="title">The Death of Polling?</span>
             <span class="author">by&nbsp;Ipsos MORI</span>
             <span class="view-count">2665&nbsp;views</span>
           </span>
         </a>
        </li>
        <li>
         <a href="/DeantaGlobal/publishing-production-from-the-desktop-to-the-cloud" title="Publishing Production: From the Des..." >
           <img class="j-thumbnail" data-original="//cdn.slidesharecdn.com/ss_thumbnails/desktoptothecloud-160226132022-thumbnail.jpg?cb=1456521576" alt="Publishing Production: From the Des..." />
           <span class="presentation-meta">
             <span class="title">Publishing Production: From the Des...</span>
             <span class="author">by&nbsp;Deanta </span>
             <span class="view-count">1777&nbsp;views</span>
           </span>
         </a>
        </li>
        <li>
         <a href="/jonoYoung/1280x800geekendprotypingfordesigners" title="Mobile App Prototyping for Designers" >
           <img class="j-thumbnail" data-original="//cdn.slidesharecdn.com/ss_thumbnails/0c780e37-2098-4e84-971b-160c56b9f3ec-160226200020-thumbnail.jpg?cb=1456521500" alt="Mobile App Prototyping for Designers" />
           <span class="presentation-meta">
             <span class="title">Mobile App Prototyping for Designers</span>
             <span class="author">by&nbsp;Jono Young</span>
             <span class="view-count">2296&nbsp;views</span>
           </span>
         </a>
        </li>
        <li>
         <a href="/visually/how-email-and-great-content-can-fuel-your-buyers-journey" title="How Email and Great Content Can Fue..." >
           <img class="j-thumbnail" data-original="//cdn.slidesharecdn.com/ss_thumbnails/visuallyemmadeck-edit-3-160226193745-thumbnail.jpg?cb=1456521484" alt="How Email and Great Content Can Fue..." />
           <span class="presentation-meta">
             <span class="title">How Email and Great Content Can Fue...</span>
             <span class="author">by&nbsp;Visually</span>
             <span class="view-count">2447&nbsp;views</span>
           </span>
         </a>
        </li>
      </ul>
    </div>

  </div> <!-- end of div class pro-overlay -->
  <div class="j-modal-share modal-share mobile-hide" style="display: none;" id="last-screen-modal-share" data-ga-track-category="" data-ga-track-action="">
  <div class="j-modal-popup modal-popup">
    <div class="j-modal-close modal-close"></div>
    <div class="modal-content-wrapper">
      <div class="j-modal-content modal-content" id="modal-content" data-slideshowid="">
        <header class="j-tabs tabs">
          <a id="button-share-tab" class="selected j-button-share-tab">Share SlideShare</a>
          <hr class="divider"/>
        </header>
        <div class="j-share-tab share-tab">
          <div>
            <ul class="j-share-social-list share-social-list" data-canonical-url="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg">
              <li class="facebook" data-network="facebook">
                <div class="social-hover">
                  <a class="share-link" rel="nofollow" data-url="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" title="Share on Facebook">Facebook</a>
                </div>
              </li>
              <li class="twitter" data-network="twitter">
                <div class="social-hover">
                  <a class="share-link" rel="nofollow"  data-url="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" data-text="TensorFlow を使った 機械学習ことはじめ (GDG京都 機械学習勉強会) #machinelearning" data-related="" data-via="SlideShare" title="Tweet on Twitter">Twitter</a>
                </div>
              </li>
              <li class="linkedin" data-network="linkedin">
                <div class="social-hover">
                  <a class="share-link" rel="nofollow" data-url="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" data-text="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会) by Toru UENOYAMA via slideshare" title="Share on Linkedin">LinkedIn</a>
                </div>
              </li>
              <li class="googleplus" data-network="googleplus">
                <div class="social-hover">
                  <a class="share-link" rel="nofollow" data-url="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg" title="Share on Google+">Google+</a>
                </div>
              </li>
            </ul>
          </div>
          <div class="section share-email">
            <span class="header">Email</span>
            <form id="share-email-form" class="j-share-email-form">
              <input id="share-email-to" data-ga="to" class="j-share-email-to j-email-clear j-share-expand-trigger" name="recipients" placeholder="Enter email addresses"></input>
              <div class="share-email-expand j-share-expand">
                <input id="share-email-name" data-ga="name" class="j-share-email-name j-email-clear" name="name" type="text" placeholder="from…"></input>
                <textarea id="share-email-msg" data-ga="msg" class="j-share-email-msg j-email-clear" name="msg" placeholder="add a message…"></textarea>
                <span class="j-email-flash email-flash"></span>
                <input id="share-email-send" data-ga="send" class="j-share-email-send button btn btn-inverse" type="submit" value="Send" />
                <div style="clear:both"></div>
              </div>
            </form>
            <div id="email-sent" class="j-email-sent section"><div>
              <span class="success-text">Email sent successfully!</span></div>
            </div>
          </div>
          <div class="j-share-embed section share-embed">
            <span class="header">Embed</span>
            <textarea id="share-embed-link" class="j-share-embed-link j-share-expand-trigger" readonly data-ga="link"></textarea>
            <div class="share-embed-options j-share-expand">
              <div class="embed-size">
                <span class="title">Size (px)</span>
                <select class="j-embed-size-picker embed-size-picker j-update-embed" id="embed-size-picker" data-ga="size-picker"></select>
              </div>
                <div class="embed-start">
                  <span class="title">Start on</span>
                  <select class="j-embed-start-picker embed-start-picker j-update-embed" id="embed-start-picker" data-ga="start-picker"></select>
                </div>
              <div class="embed-show-related" style="display:none">
                <input type="checkbox" name="related-content" checked="checked" class="j-embed-related-cbox embed-related-cbox j-update-embed" data-ga="related">
                <span>Show related SlideShares at end</span>
              </div>
            </div>
          </div>
          <div class="wordpress-container section">
            <span class="header">WordPress Shortcode</span>
            <input type="text" name="embed-code" id="share-embed-wp" value="" readonly="readonly" class="j-share-embed-wp text quiet h-wpembedcode j-share-expand-trigger" data-ga="wp-link">
          </div>
          <div class="share-link-container section">
            <span class="header">Link</span>
            <input type="input" class="j-share-link-url j-share-expand-trigger" id="share-link-url" data-ga="link" readonly></input>
          </div>
        </div>
      </div>
    </div>
  </div>
</div><!-- share modal -->

</div><!-- last screen ends here -->

        </div>
      </div>

      <div class="slideshow-info-container" itemscope itemtype="http://schema.org/MediaObject">


        <div class="slideshow-info">
          <meta itemprop="inLanguage" content="ja">
          <meta itemprop="image" content="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-1-638.jpg?cb=1456578209">
          <meta itemprop="thumbnailUrl" content="http://cdn.slidesharecdn.com/ss_thumbnails/machinelearningwithtensorflow20160227-160227125739-thumbnail.jpg?cb=1456578209">
          <meta itemprop="embedURL" content="https://www.slideshare.net/slideshow/embed_code/key/tkPJMfS5uuYI0N">
          <meta itemprop="playerType" content="HTML5">
          <meta itemprop="interactionCount" content="UserComments:0">
          <meta itemprop="interactionCount" content="UserLikes:6">
          <meta itemprop="interactionCount" content="UserDownloads:4">
          <meta itemprop="interactionCount" content="UserPageVisits:327">
          <meta itemprop="interactionCount" content="UserPlays:327">
          <meta itemprop="interactionCount" content="UserPlusOnes:0" id="meta-google">
          <meta itemprop="interactionCount" content="UserTweets:0" id="meta-twitter">


          <div class="slideshow-title-container row add-padding-right">
            <div class="small-10 columns">
              <h1 class="notranslate slideshow-title-text" itemprop="headline">
                <span class="j-title-breadcrumb">
                  TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会)
                </span>
              </h1>
            </div>

            <div class="small-2 columns text-right format-views" data-views="views">
                <span class="notranslate">
                  327<br>
                </span>
                <span class="pippin-data hide">  -1</span>
            </div>
          </div>

          <ul id="slideshow-actions" class="slideshow-actions">
            <li class="item-action">
              <button class="tiny art-deco share" data-action="share">Share</button>
            </li>
              <li class="item-action">
                <button class="tiny art-deco like button" data-action="like" href="/signup?login_source=slideview.popup.like&amp;from=favorite&amp;layout=foundation&amp;from_source=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg" rel="nofollow">Like</button>
              </li>
              <li class="item-action">
                <button class="tiny art-deco download button" data-action="download" href="/login?from_source=%2FToruUenoyama%2Ftensorflow-gdg%3Ffrom_action%3Dsave&amp;from=download&amp;layout=foundation" rel="nofollow">
                  Download
                </button>
              </li>
          </ul>


          <div class="author-container add-padding-right" itemprop="author" itemscope itemtype="http://schema.org/Person">

            <div class="left author-thumbnail">
              <a href="/ToruUenoyama?utm_campaign=profiletracking&amp;utm_medium=sssite&amp;utm_source=ssslideview" class="author-photo-wrapper" title="ToruUenoyama" itemprop="url">
                <img alt="Toru UENOYAMA" class="author-photo" itemprop="image" src="//cdn.slidesharecdn.com/profile-photo-ToruUenoyama-48x48.jpg?cb=1456615377" />
              </a>
            </div>

            <div class="author-text">
              <h2 style="display:inline;">
                <a class="j-author-name"
                  title="ToruUenoyama"
                  rel="author"
                  href="/ToruUenoyama?utm_campaign=profiletracking&amp;utm_medium=sssite&amp;utm_source=ssslideview"
                  data-ga-cat="bigfoot_slideview"
                  data-ga-action="authorlinkclick">
                  <span itemprop="name">Toru UENOYAMA</span></a></h2><small class="lighter-color-text" itemprop="jobTitle">, Software Engineer</small>


              <div class="author-cta-container">

                  <div class="follow-container">
                    <span class="j-follow " data-contactee-id="32207858">

                        <a class="follow-btn" data-contactee="32207858" href="/signup?login_source=slideview.popup.follow&from=addcontact&from_source=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg">
                          <i class="fa fa-plus"></i> Follow
                        </a>

                      <div class="j-follow-progress indicator">
                        <i class="fa fa-spinner fa-spin"></i>
                      </div>
                    </span>
                  </div>
              </div>
            </div>
          </div>

            <div class="social-share-container add-padding-right">
              <button id="sv-linkedin-share" class="j-share-item button-li tiny radius" data-network="linkedin" data-ga-action="viralshareLinkedIn_click">
                <i class="fa fa-linkedin fa-lg"></i>
                <span class="separator"></span>
                <span class="j-share-count share-count">0</span>
              </button>

              <button id="sv-facebook-share" class="j-share-item button-fb tiny radius" data-network="facebook" data-ga-action="viralsharefacebook">
                <i class="fa fa-facebook fa-lg"></i>
                <span class="separator"></span>
                <span class="j-share-count share-count">0</span>
              </button>

              <button id="sv-twitter-share" class="j-share-item button-tw tiny radius" data-network="twitter" data-ga-action="viralsharetwitter_click">
                <i class="fa fa-twitter fa-lg"></i>
                <span class="separator"></span>
                <span class="j-share-count share-count">0</span>
              </button>

              <button id="sv-google-share" class="j-share-item button-go tiny radius" data-network="google" data-ga-action="viralshareGooglePlusOne">
                <i class="fa fa-google-plus fa-lg"></i>
                <span class="separator"></span>
                <span class="j-share-count share-count">0</span>
              </button>
            </div>

          <p>
            <small>
              Published on <time datetime="Feb 27, 2016" itemprop="datePublished">Feb 27, 2016</time>
            </small>
          </p>

            <div class="slideshow-description-container add-padding-right">
              <div class="description row"
                data-ga-cat="bigfoot_slideview"
                data-ga-action="description>more">
                <div class="large-10 columns">
                  <p id="slideshow-description-paragraph" class="notranslate">
                    TensorFlow の基本的な使い方やコーディング方法を、線形重回帰やロジスティック回帰、パーセプトロンの実装例を通じて解説しています。（機械学習初心者向け）<br /><br />※2016&#x2F;02&#x2F;27に開催した「GDG京都 機械学習勉強会」で発表した資料です。
                  </p>
                </div>
                <div class="large-2 columns">
                  <button class="j-expand-text empty_btn_design">
                    ...<i class="fa fa-caret-down"></i>
                  </button>
                </div>
              </div>
            </div>


            <div class="categories-container add-padding-right">
              <span>Published in:</span>
                <a rel="nofollow" href="/featured/category/technology">Technology</a><span class="comma"></span>
            </div>



        </div>


        <div class="slideshow-tabs-container show-for-medium-up">
          <dl class="tabs" data-tab>
              <dd class="active">
                <a href="#comments-panel">
                  <i class="fa fa-comment"></i>
                    0 Comments
                </a>
              </dd>

            <dd class="">
              <a href="#likes-panel">
                <i class="fa fa-heart"></i>
                <span class="j-favs-count">
                  6 Likes
                </span>
              </a>
            </dd>


            <dd>
              <a href="#stats-panel" class="j-stats-tab">
                <i class="fa fa-bar-chart"></i>
                Statistics
              </a>
            </dd>

            <dd>
              <a href="#notes-panel">
                <i class="fa fa-file-text"></i>
                Notes
              </a>
            </dd>
          </dl>

          <div class="tabs-content">

              <div id="comments-panel" class="content active commentsWrapper commentsNotes">
                <ul class="hide">

                  <li id="commentsTemplate">
                    <div class="row">
                      <div class="small-1 columns thumbnail">
                        <a class="j-author-photo notranslate commenter" title="Commenter Title" rel="nofollow">
                          <img class="nickname"
                            alt="Full Name"
                            src="//public.slidesharecdn.com/b/images/user-48x48.png" />
                        </a>
                      </div>

                      <div class="small-11 columns">
                        <a class="j-author-photo notranslate commenter" title="Commenter Title" rel="nofollow">
                          <span class="j-username notranslate" data-ga-cat="bigfoot_slideview" data-ga-action="commentuserlinkclick">Full Name</span>
                          <span class="bioStub notranslate">
                            <span class="j-commenter-role"></span>
                            <span class="j-commenter-org"></span>
                          </span>
                        </a>

                        <div class="commentText notranslate">
                          Comment goes here.
                        </div>

                        <time class="commentTimestamp small-text lighter-color-text">12 hours ago</time>&nbsp;&nbsp;

                        <span class="commentMeta">
                          <span class="commentActions small-text">
                            <a href="#" class="j-action-delete">Delete</a>
                            <a href="#" class="j-reply">Reply</a>
                            <a href="#" class="j-action-spam">Spam</a>
                            <a href="#" class="j-action-block">Block</a>
                          </span>
                        </span>

                        <div id="confirmDialog" class="panel callout block-message">
                          <div>
                            <span class="title">Are you sure you want to</span>
                            <a href="#" id="yes">Yes</a>
                            <a href="#" id="no">No</a>
                          </div>
                        </div>

                        <div id="messageDialog" class="block-message">
                          Your message goes here
                        </div>

                        <span class="j-loading" style="display: none;">
                          <i class="fa fa-spinner fa-spin"></i>
                        </span>
                      </div>
                    </div>
                  </li>
                </ul>


                  <form action="#" method="post" accept-charset="utf-8" id="postComment" class="j-comment-post postComment">
                    <div class="row">
                      <div class="small-1 columns thumbnail">
                          <div class="left">
                            <img class="nickname" alt="no profile picture user"
                              src="//public.slidesharecdn.com/b/images/user-48x48.png" />
                          </div>
                      </div>

                      <div class="small-11 columns">
                        <div class="row collapse">
                          <div class="small-10 columns">
                            <input class="j-post-comment-input comment-text" type="text" placeholder="Share your thoughts…" />
                          </div>
                          <div class="small-2 columns">
                              <a id="login-provider-slideshare" class="postfix" rel="nofollow" href="/signup?login_source=slideview.popup.comment&from=comments&from_source=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg">
                                <button type="button" class="postfix">Post</button>
                              </a>
                          </div>
                        </div>
                      </div>
                    </div>
                  </form>

                <ul id="commentsList" class="user-list no-bullet">
                    <li>
                      <p class="empty-stat-box text-center">
                        <em>Be the first to comment</em>
                      </p>
                    </li>
                </ul>

              </div>


            <div class="content" id="likes-panel">
              <ul id="favsList" class="j-favs-list notranslate user-list no-bullet" itemtype="http://schema.org/UserLikes" itemscope>
                    <li itemtype="http://schema.org/Person" itemscope>
                      <div class="row">

                        <div class="small-1 columns thumbnail">
                          <a class="j-author-photo notranslate"
                            title="cm3"
                            itemprop="url"
                            rel="nofollow"
                            href="/cm3?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <img itemprop="image"
                              class="j-lazy-thumb"
                              alt="cm3"
                              src="//public.slidesharecdn.com/b/images/user-48x48.png"
                              data-original="//cdn.slidesharecdn.com/profile-photo-cm3-48x48.jpg" />
                          </a>
                        </div>

                        <div class="small-11 columns">

                          <a class="favoriter notranslate"
                            title="cm3"
                            rel="nofollow"
                            href="/cm3?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <span class="j-username notranslate" data-ga-cat="bigfoot_slideview" data-ga-action="favoriteuserlinkclick" itemprop="name">Akihiro Kameda</span>
                            <span class="bioStub notranslate small-text light-color-text">
                              <span class="j-favoriter-role">
                                
                                <span></span>
                              </span>
                              <span class="j-favoriter-org">
                                 at 
                                <span>Kyoto University</span>
                              </span>
                            </span>

                            <div class="j-tags favTags"></div>

                            <time class="commentTimestamp small-text lighter-color-text">
                              4 hours ago
                            </time>
                          </a>
                        </div>
                      </div>
                    </li>
                    <li itemtype="http://schema.org/Person" itemscope>
                      <div class="row">

                        <div class="small-1 columns thumbnail">
                          <a class="j-author-photo notranslate"
                            title="yasutomo57jp"
                            itemprop="url"
                            rel="nofollow"
                            href="/yasutomo57jp?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <img itemprop="image"
                              class="j-lazy-thumb"
                              alt="yasutomo57jp"
                              src="//public.slidesharecdn.com/b/images/user-48x48.png"
                              data-original="//cdn.slidesharecdn.com/profile-photo-yasutomo57jp-48x48.jpg" />
                          </a>
                        </div>

                        <div class="small-11 columns">

                          <a class="favoriter notranslate"
                            title="yasutomo57jp"
                            rel="nofollow"
                            href="/yasutomo57jp?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <span class="j-username notranslate" data-ga-cat="bigfoot_slideview" data-ga-action="favoriteuserlinkclick" itemprop="name">Yasutomo Kawanishi</span>
                            <span class="bioStub notranslate small-text light-color-text">
                              <span class="j-favoriter-role">
                                , 
                                <span>Assistant Professor - Nagoya University</span>
                              </span>
                              <span class="j-favoriter-org">
                                 at 
                                <span>Nagoya University</span>
                              </span>
                            </span>

                            <div class="j-tags favTags"></div>

                            <time class="commentTimestamp small-text lighter-color-text">
                              8 hours ago
                            </time>
                          </a>
                        </div>
                      </div>
                    </li>
                    <li itemtype="http://schema.org/Person" itemscope>
                      <div class="row">

                        <div class="small-1 columns thumbnail">
                          <a class="j-author-photo notranslate"
                            title="yijiri"
                            itemprop="url"
                            rel="nofollow"
                            href="/yijiri?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <img itemprop="image"
                              class="j-lazy-thumb"
                              alt="yijiri"
                              src="//public.slidesharecdn.com/b/images/user-48x48.png"
                              data-original="//cdn.slidesharecdn.com/profile-photo-yijiri-48x48.jpg" />
                          </a>
                        </div>

                        <div class="small-11 columns">

                          <a class="favoriter notranslate"
                            title="yijiri"
                            rel="nofollow"
                            href="/yijiri?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <span class="j-username notranslate" data-ga-cat="bigfoot_slideview" data-ga-action="favoriteuserlinkclick" itemprop="name">Yoshihisa Ijiri</span>
                            <span class="bioStub notranslate small-text light-color-text">
                              <span class="j-favoriter-role">
                                , 
                                <span>Engineer at OMRON Corporation</span>
                              </span>
                              <span class="j-favoriter-org">
                                 at 
                                <span>OMRON Corporation</span>
                              </span>
                            </span>

                            <div class="j-tags favTags"></div>

                            <time class="commentTimestamp small-text lighter-color-text">
                              9 hours ago
                            </time>
                          </a>
                        </div>
                      </div>
                    </li>
                    <li itemtype="http://schema.org/Person" itemscope>
                      <div class="row">

                        <div class="small-1 columns thumbnail">
                          <a class="j-author-photo notranslate"
                            title="yuzataku"
                            itemprop="url"
                            rel="nofollow"
                            href="/yuzataku?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <img itemprop="image"
                              class="j-lazy-thumb"
                              alt="yuzataku"
                              src="//public.slidesharecdn.com/b/images/user-48x48.png"
                              data-original="//cdn.slidesharecdn.com/profile-photo-yuzataku-48x48.jpg" />
                          </a>
                        </div>

                        <div class="small-11 columns">

                          <a class="favoriter notranslate"
                            title="yuzataku"
                            rel="nofollow"
                            href="/yuzataku?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <span class="j-username notranslate" data-ga-cat="bigfoot_slideview" data-ga-action="favoriteuserlinkclick" itemprop="name">yuzataku</span>
                            <span class="bioStub notranslate small-text light-color-text">
                              <span class="j-favoriter-role">
                                
                                <span></span>
                              </span>
                              <span class="j-favoriter-org">
                                
                                <span></span>
                              </span>
                            </span>

                            <div class="j-tags favTags"></div>

                            <time class="commentTimestamp small-text lighter-color-text">
                              9 hours ago
                            </time>
                          </a>
                        </div>
                      </div>
                    </li>
                    <li itemtype="http://schema.org/Person" itemscope>
                      <div class="row">

                        <div class="small-1 columns thumbnail">
                          <a class="j-author-photo notranslate"
                            title="kumadasu"
                            itemprop="url"
                            rel="nofollow"
                            href="/kumadasu?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <img itemprop="image"
                              class="j-lazy-thumb"
                              alt="kumadasu"
                              src="//public.slidesharecdn.com/b/images/user-48x48.png"
                              data-original="//cdn.slidesharecdn.com/profile-photo-kumadasu-48x48.jpg" />
                          </a>
                        </div>

                        <div class="small-11 columns">

                          <a class="favoriter notranslate"
                            title="kumadasu"
                            rel="nofollow"
                            href="/kumadasu?utm_campaign=profiletracking&utm_medium=sssite&utm_source=ssslideshow">

                            <span class="j-username notranslate" data-ga-cat="bigfoot_slideview" data-ga-action="favoriteuserlinkclick" itemprop="name">Hiroshi Sakate</span>
                            <span class="bioStub notranslate small-text light-color-text">
                              <span class="j-favoriter-role">
                                , 
                                <span>at</span>
                              </span>
                              <span class="j-favoriter-org">
                                 at 
                                <span>Fujitsu Ten Limited</span>
                              </span>
                            </span>

                            <div class="j-tags favTags"></div>

                            <time class="commentTimestamp small-text lighter-color-text">
                              11 hours ago
                            </time>
                          </a>
                        </div>
                      </div>
                    </li>
              </ul>

                <div class="more-container text-center">
                  <a href="#" class="j-more-favs">
                    Show More
                    <div style="inline-block"><i class="fa fa-caret-down"></i></div>
                  </a>
                </div>
            </div>

            <div class="content" id="downloads-panel">
                <div class="empty-stat-box">No Downloads</div>
            </div>

            <div class="content" id="stats-panel">
              <div class="row info-stats">
                <div class="small-4 columns">
                  <strong>Views</strong>
                  <div class="row">
                    <div class="small-8 columns stat-label">Total views</div>
                    <div class="small-4 columns stat-value text-right">
                      327
                    </div>

                    <div class="small-8 columns stat-label">On SlideShare</div>
                    <div class="j-slideshare-views small-4 columns stat-value text-right ">
                      0
                    </div>

                    <div class="small-8 columns stat-label">From Embeds</div>
                    <div class="j-embed-views small-4 columns stat-value text-right">
                      0
                    </div>

                    <div class="small-8 columns stat-label">Number of Embeds</div>
                    <div class="small-4 columns stat-value text-right">
                      0
                    </div>
                  </div>
                </div>

                <div class="small-4 columns">
                  <strong>Actions</strong>
                  <div class="row">
                    <div class="small-8 columns stat-label">Shares</div>
                    <div class="small-4 columns stat-value text-right j-total-shares">0</div>

                    <div class="small-8 columns stat-label">Downloads</div>
                    <div class="small-4 columns stat-value text-right ">
                      4
                    </div>

                    <div class="small-8 columns stat-label">Comments</div>
                    <div class="small-4 columns stat-value text-right">
                      0
                    </div>

                    <div class="small-8 columns stat-label">Likes</div>
                    <div class="small-4 columns stat-value text-right">
                      6
                    </div>
                  </div>
                </div>

                <div class="small-4 columns">
                  <strong>
                    Embeds
                    <span class="j-embed-views notranslate from-embed hint">0</span>
                  </strong>

                  <div class="j-info-embeds">
                      <div class="j-no-embeds no-embeds">No embeds</div>
                    <div class="row no-translate j-embeds-container" style="max-height:120px; overflow:auto;">
                    </div>
                  </div>
                </div>
              </div>

              <hr>

              <div class="row">
                <div class="small-12 columns">
                  <strong class="copy-in-aria-label" aria-label="Report content"></strong><br>
                  <div class="flag flag-inappropriate">
                    <a class="action-flag" rel="nofollow" href="/signup?login_source=slideview.popup.flags&amp;from=flagss&amp;from_source=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg">
                      <span class="j-tooltip flagged copy-in-aria-label" title="This Presentation has been flagged" aria-label="Flagged as inappropriate"></span>
                      <span class="j-tooltip flag copy-in-aria-label" title="Flag this presentation as inappropriate" aria-label="Flag as inappropriate"></span>
                    </a>
                  </div>

                  <div>
                    <a href="http://www.linkedin.com/legal/copyright-policy" rel="nofollow" class="copy-in-aria-label" aria-label="Copyright Complaint"></a>
                  </div>
                </div>
              </div>
            </div>

            <div class="content" id="notes-panel">
              <div id="empty-note" class="empty-stat-box">No notes for slide</div>
            </div>
          </div>
        </div>

            <div class="notranslate transcript add-padding-right j-transcript"  >
              <h3 class="transcript-header">
                  <i class="fa fa-file-o"></i>
                  TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会)
              </h3>
              <ol class="j-transcripts transcripts no-bullet no-style" itemprop="text">
                    <li>
      1.
    TensorFlow を使った
機械学習ことはじめ
2016-02-27
GDG京都 上野山
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-2-638.jpg?cb=1456578209" title="自己紹介
 電機メーカ勤務の
似非ソフト屋
 もともと組込み
 画像処理
 データ解析 と変遷
 最近は A..." target="_blank">
        2.
      </a>
    自己紹介
 電機メーカ勤務の
似非ソフト屋
 もともと組込み
 画像処理
 データ解析 と変遷
 最近は Android より
R言語(統計処理言語)と
格闘中
2
上野山 徹
うえのやま とおる
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-3-638.jpg?cb=1456578209" title="1. TensorFlow とは
2016-02-27 GDG京都 機械学習勉強会
3
 " target="_blank">
        3.
      </a>
    1. TensorFlow とは
2016-02-27 GDG京都 機械学習勉強会
3
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-4-638.jpg?cb=1456578209" title="TensorFlow とは
 Google社のMachine Intelligence
研究所が開発した数値計算ライ..." target="_blank">
        4.
      </a>
    TensorFlow とは
 Google社のMachine Intelligence
研究所が開発した数値計算ライブラリ
 Deep Learning をはじめとする 機械学習 の
研究や活用にGoogle社内で用いられている
 2011年から使用していたDistBeliefを強化
 複数のCPU/GPUや、複数のマシンに分散して
実行できる。(※複数マシン対応版は未公開)
 C++とPython 2or3 APIを持っている
 2015年11月にオープンソース(Apache2.0)公開
2016-02-27 GDG京都 機械学習勉強会
4
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-5-638.jpg?cb=1456578209" title="データフローグラフ
2016-02-27 GDG京都 機械学習勉強会
5
 掛け算
 足し算
 関数
テンソル ..." target="_blank">
        5.
      </a>
    データフローグラフ
2016-02-27 GDG京都 機械学習勉強会
5
 掛け算
 足し算
 関数
テンソル が データフローグラフ 上を
流れるように計算処理している
テンソル 
(多次元配列)
𝐶 = … (𝑅𝑒𝐿𝑈 𝑏 + 𝑊𝑋 )
 数式で表現すると、
TensorFlow
巨大なデータにも使える
行列計算ライブラリ
ビッグデータの
機械学習に使える！出典）TensorFlow: Large-Scale Machine Learning
on Heterogeneous Distributed Systems, Google, 2015
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-6-638.jpg?cb=1456578209" title="TensorBoard
 データフローグラフのネットワーク構造や
学習経過をWeb上で確認できる
2016-02-2..." target="_blank">
        6.
      </a>
    TensorBoard
 データフローグラフのネットワーク構造や
学習経過をWeb上で確認できる
2016-02-27 GDG京都 機械学習勉強会
6
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-7-638.jpg?cb=1456578209" title="詳細は Web へ
2016-02-27 GDG京都 機械学習勉強会
7
https://speakerdeck.co..." target="_blank">
        7.
      </a>
    詳細は Web へ
2016-02-27 GDG京都 機械学習勉強会
7
https://speakerdeck.com/kazunori279/cloud-vision-api-and-tensorflow
例えば、佐藤さん@Google の資料など
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-8-638.jpg?cb=1456578209" title="2. TensorFlow を使った
機械学習
2016-02-27 GDG京都 機械学習勉強会
8
 " target="_blank">
        8.
      </a>
    2. TensorFlow を使った
機械学習
2016-02-27 GDG京都 機械学習勉強会
8
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-9-638.jpg?cb=1456578209" title="さっそく問題です
2016-02-27 GDG京都 機械学習勉強会
9
たかしくんは八百屋へお使いに行きました。
リン..." target="_blank">
        9.
      </a>
    さっそく問題です
2016-02-27 GDG京都 機械学習勉強会
9
たかしくんは八百屋へお使いに行きました。
リンゴ１個とミカン３個を買うと１９０円，
リンゴ３個とミカン１個を買うと３３０円
するようです。
リンゴ２個とミカン４個を買うと
いくらになるでしょうか？
例題１
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-10-638.jpg?cb=1456578209" title="円
式で表すと...
2016-02-27 GDG京都 機械学習勉強会
10
円 × 1 ＋ 円 × 3 ＝ 190 ..." target="_blank">
        10.
      </a>
    円
式で表すと...
2016-02-27 GDG京都 機械学習勉強会
10
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-11-638.jpg?cb=1456578209" title="円
回答
2016-02-27 GDG京都 機械学習勉強会
11
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × ..." target="_blank">
        11.
      </a>
    円
回答
2016-02-27 GDG京都 機械学習勉強会
11
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
100 30
100 30
100 30
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-12-638.jpg?cb=1456578209" title="円
回答
2016-02-27 GDG京都 機械学習勉強会
12
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × ..." target="_blank">
        12.
      </a>
    円
回答
2016-02-27 GDG京都 機械学習勉強会
12
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ 320
100 30
100 30
100 30
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-13-638.jpg?cb=1456578209" title="円
これも機械学習（回帰）の問題です！
2016-02-27 GDG京都 機械学習勉強会
13
円 × 1 ＋ 円 ×..." target="_blank">
        13.
      </a>
    円
これも機械学習（回帰）の問題です！
2016-02-27 GDG京都 機械学習勉強会
13
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ 320
100 30
100 30
100 30
学習
予測
実績データからパラメータを推定
パラメータを使って結果を予測
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-14-638.jpg?cb=1456578209" title="実は店のおやじは気まぐれでした
2016-02-27 GDG京都 機械学習勉強会
14
円
円 × 1 ＋ 円 × 3..." target="_blank">
        14.
      </a>
    実は店のおやじは気まぐれでした
2016-02-27 GDG京都 機械学習勉強会
14
円
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
円 × 5 ＋ 円 × 7 ＝ 660 円
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-15-638.jpg?cb=1456578209" title="実は店のおやじは気まぐれでした
2016-02-27 GDG京都 機械学習勉強会
15
円
円 × 1 ＋ 円 × 3..." target="_blank">
        15.
      </a>
    実は店のおやじは気まぐれでした
2016-02-27 GDG京都 機械学習勉強会
15
円
円 × 1 ＋ 円 × 3 ＝ 190 円
円 × 3 ＋ 円 × 1 ＝ 330 円
円 × 2 ＋ 円 × 4 ＝ ?
円 × 5 ＋ 円 × 7 ＝ 660 円
100 30
100 30
100 30
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-16-638.jpg?cb=1456578209" title="どうやってパラメータ(単価)を見つけるか？
2016-02-27 GDG京都 機械学習勉強会
16
考え方
合計金額の..." target="_blank">
        16.
      </a>
    どうやってパラメータ(単価)を見つけるか？
2016-02-27 GDG京都 機械学習勉強会
16
考え方
合計金額の誤差が小さくなるような
単価を採用しよう!
円 × 1 ＋ 円 × 3  40円 - 190 円
円 × 3 ＋ 円 × 1  40円 - 330 円
円 × 5 ＋ 円 × 7  120円 - 660 円
10 10
10 10
10 10
合計金額
（正解）
合計金額
（予測）
(２乗)誤差
 22500
 84100
 291600
( )2
( )2
( )2
平均 132733
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-17-638.jpg?cb=1456578209" title="すべての組合せを計算
2016-02-27 GDG京都 機械学習勉強会
17
ミカン
0 10 20 30 40 50..." target="_blank">
        17.
      </a>
    すべての組合せを計算
2016-02-27 GDG京都 機械学習勉強会
17
ミカン
0 10 20 30 40 50 60 70 80 90 100
リンゴ
0 193533 158700 127800 100833 77800 58700 43533 32300 25000 21633 22200
10 164833 132733 104567 80333 60033 43667 31233 22733 18167 17533 20833
20 138467 109100 83667 62167 44600 30967 21267 15500 13667 15767 21800
30 114433 87800 65100 46333 31500 20600 13633 10600 11500 16333 25100
40 92733 68833 48867 32833 20733 12567 8333 8033 11667 19233 30733
50 73367 52200 34967 21667 12300 6867 5367 7800 14167 24467 38700
60 56333 37900 23400 12833 6200 3500 4733 9900 19000 32033 49000
70 41633 25933 14167 6333 2433 2467 6433 14333 26167 41933 61633
80 29267 16300 7267 2167 1000 3767 10467 21100 35667 54167 76600
90 19233 9000 2700 333 1900 7400 16833 30200 47500 68733 93900
100 11533 4033 467 833 5133 13367 25533 41633 61667 85633 113533
110 6167 1400 567 3667 10700 21667 36567 55400 78167 104867 135500
120 3133 1100 3000 8833 18600 32300 49933 71500 97000 126433 159800
130 2433 3133 7767 16333 28833 45267 65633 89933 118167 150333 186433
140 4067 7500 14867 26167 41400 60567 83667 110700 141667 176567 215400
150 8033 14200 24300 38333 56300 78200 104033 133800 167500 205133 246700
近そうな値は見つかるけど、計算は大変！
（もし単価が0円～1万円で、10種類買ったら?）
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-18-638.jpg?cb=1456578209" title="よく見ると、徐々に変化している
2016-02-27 GDG京都 機械学習勉強会
18
ミカン
0 10 20 30 ..." target="_blank">
        18.
      </a>
    よく見ると、徐々に変化している
2016-02-27 GDG京都 機械学習勉強会
18
ミカン
0 10 20 30 40 50 60 70 80 90 100
リンゴ
0 193533 158700 127800 100833 77800 58700 43533 32300 25000 21633 22200
10 164833 132733 104567 80333 60033 43667 31233 22733 18167 17533 20833
20 138467 109100 83667 62167 44600 30967 21267 15500 13667 15767 21800
30 114433 87800 65100 46333 31500 20600 13633 10600 11500 16333 25100
40 92733 68833 48867 32833 20733 12567 8333 8033 11667 19233 30733
50 73367 52200 34967 21667 12300 6867 5367 7800 14167 24467 38700
60 56333 37900 23400 12833 6200 3500 4733 9900 19000 32033 49000
70 41633 25933 14167 6333 2433 2467 6433 14333 26167 41933 61633
80 29267 16300 7267 2167 1000 3767 10467 21100 35667 54167 76600
90 19233 9000 2700 333 1900 7400 16833 30200 47500 68733 93900
100 11533 4033 467 833 5133 13367 25533 41633 61667 85633 113533
110 6167 1400 567 3667 10700 21667 36567 55400 78167 104867 135500
120 3133 1100 3000 8833 18600 32300 49933 71500 97000 126433 159800
130 2433 3133 7767 16333 28833 45267 65633 89933 118167 150333 186433
140 4067 7500 14867 26167 41400 60567 83667 110700 141667 176567 215400
150 8033 14200 24300 38333 56300 78200 104033 133800 167500 205133 246700
ミカン 0
ミカン 40
ミカン 800
50000
100000
150000
200000
250000
0 20 40 60 80
100120
140
リンゴ
最小値
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-19-638.jpg?cb=1456578209" title="小さくなる方向へたどっていくと...
2016-02-27 GDG京都 機械学習勉強会
19
ミカン
0 10 20 ..." target="_blank">
        19.
      </a>
    小さくなる方向へたどっていくと...
2016-02-27 GDG京都 機械学習勉強会
19
ミカン
0 10 20 30 40 50 60 70 80 90 100
リンゴ
0 193533
10 132733
20 83667
30 46333
40 20733
50 6867
60 3500
70 2467
80 7267 2167 1000
90 2700 333 1900
100 467 833 5133
110
120
130
140
150
ミカン 0
ミカン 40
ミカン 800
50000
100000
150000
200000
250000
0 20 40 60 80
100120
140
リンゴ
最小値
少ない計算で早く最小値に！
※最小(大)値を求める処理＝&quot;最適化&quot;
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-20-638.jpg?cb=1456578209" title="気まぐれおやじのこころを読む
2016-02-27 GDG京都 機械学習勉強会
20
円
円 × 1 ＋ 円 × 3 ..." target="_blank">
        20.
      </a>
    気まぐれおやじのこころを読む
2016-02-27 GDG京都 機械学習勉強会
20
円
円 × 1 ＋ 円 × 3 ≒ 190 円
円 × 3 ＋ 円 × 1 ≒ 330 円
円 × 2 ＋ 円 × 4 ≒
円 × 5 ＋ 円 × 7 ≒ 660 円
90 30
90 30
90 30
90 30 310
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-21-638.jpg?cb=1456578209" title="ここまでの振り返り
2016-02-27 GDG京都 機械学習勉強会
21
TensorFlowでも同じように書いてい..." target="_blank">
        21.
      </a>
    ここまでの振り返り
2016-02-27 GDG京都 機械学習勉強会
21
TensorFlowでも同じように書いていく
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-22-638.jpg?cb=1456578209" title="線形回帰(線形重回帰)の1つ
2016-02-27 GDG京都 機械学習勉強会
22
入力層 出力層
𝑦
𝑥1
𝑥2
..." target="_blank">
        22.
      </a>
    線形回帰(線形重回帰)の1つ
2016-02-27 GDG京都 機械学習勉強会
22
入力層 出力層
𝑦
𝑥1
𝑥2
× 𝑎1
𝑦 = 𝑎1 𝑥1 + 𝑎2 𝑥2
× 𝑎2
リンゴの数 
ミカンの数 
合計金額
リンゴ
の単価
↓
↑
ミカン
の単価
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-23-638.jpg?cb=1456578209" title="TensorFlow の処理の流れ
2016-02-27 GDG京都 機械学習勉強会
23
# coding: utf..." target="_blank">
        23.
      </a>
    TensorFlow の処理の流れ
2016-02-27 GDG京都 機械学習勉強会
23
# coding: utf-8
# 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
# 1. 予測式(モデル)を記述する
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=(None, 2), name=&quot;x&quot;)
y_ = tf.placeholder(tf.float32, shape=(None, 1), name=&quot;y&quot;)
# モデルパラメータ
a = tf.Variable(tf.zeros((2, 1)), name=&quot;a&quot;)
# モデル式
y = tf.matmul(x, a)
# 2. 学習に必要な関数を定義する
# 誤差関数(loss)
loss = tf.reduce_mean(tf.square(y_ - y))
# 最適化手段を選ぶ(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.02).minimize(loss)
# 3. 実際に学習処理を実行する
# (1) 訓練データを生成する
train_x = np.array([[1., 3.], [3., 1.], [5., 7.]])
train_y = np.array([190., 330., 660.]).reshape(3, 1)
print &quot;x=&quot;, train_x
print &quot;y=&quot;, train_y
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
# (3) 最急勾配法でパラメータ更新 (100回更新する)
for i in range(100):
_, l, a_ = sess.run([train_step, loss, a], feed_dict={x: train_x, y_: train_y})
if (i + 1) % 10 == 0:
print &quot;step=%3d, a1=%6.2f, a2=%6.2f, loss=%.2f&quot; % (i + 1, a_[0], a_[1], l)
# (4) 学習結果を出力
est_a = sess.run(a, feed_dict={x: train_x, y_: train_y})
print &quot;Estimated: a1=%6.2f, a2=%6.2f&quot; % (est_a[0], est_a[1])
# 4. 新しいデータに対して予測する
# (1) 新しいデータを用意
new_x = np.array([2., 4.]).reshape(1, 2)
# (2) 学習結果をつかって，予測実施
new_y = sess.run(y, feed_dict={x: new_x})
print new_y
# 5. 後片付け
# セッションを閉じる
sess.close()
0. ライブラリをロード
1. 予測式(モデル)を記述
2. 誤差関数と最適化手法を記述
3. 訓練データを作成(or読込)し
学習実行
4. 予測
5. 後片付け
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-24-638.jpg?cb=1456578209" title="0. ライブラリをロード
2016-02-27 GDG京都 機械学習勉強会
24
import numpy as np..." target="_blank">
        24.
      </a>
    0. ライブラリをロード
2016-02-27 GDG京都 機械学習勉強会
24
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
TensorFlow のほか
高速な配列演算パッケージであるNumPyや
プロットしたい場合はmatplotlib.pyplotなど
使用するライブラリをロードする
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-25-638.jpg?cb=1456578209" title="1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
25
# 入力変数と出力変数のプレース..." target="_blank">
        25.
      </a>
    1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
25
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=(None, 2), name=&quot;x&quot;)
y_ = tf.placeholder(tf.float32, shape=(None, 1), name=&quot;y&quot;)
# モデルパラメータ
a = tf.Variable(tf.zeros((2, 1)), name=&quot;a&quot;)
# モデル式
y = tf.matmul(x, a)
入出力変数  placeholder, パラメータ  Variable
・値を入れる&quot;箱&quot;を作成
（注意）shape = (訓練データの数, 次元数)
※&quot;訓練データの数&quot;は None にすると可変長扱い
・予測式(モデル)を記述
y x a
190
330
660
1 3
3 1
5 7
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-26-638.jpg?cb=1456578209" title="2.誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
26
# 誤差関数(loss)
lo..." target="_blank">
        26.
      </a>
    2.誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
26
# 誤差関数(loss)
loss = tf.reduce_mean(tf.square(y_ - y))
# 最適化手段を選ぶ(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.02).minimize(loss)
・誤差関数を記述
ふつうの回帰問題  平均二乗誤差
1
𝑁
𝑦𝑖 − 𝑦𝑖
2
𝑁
𝑖
・最適化手法を選ぶ
・入門  最急降下法(勾配降下法) GradientDescent～
※どれを選ぶかで学習(最適化)の速さが変わる
・引数に適度な&quot;学習率&quot;を指定する
※大きすぎると学習失敗(発散), 小さすぎると学習が遅い
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-27-638.jpg?cb=1456578209" title="3. 訓練データを作成(or読込)し，
2016-02-27 GDG京都 機械学習勉強会
27
train_x = n..." target="_blank">
        27.
      </a>
    3. 訓練データを作成(or読込)し，
2016-02-27 GDG京都 機械学習勉強会
27
train_x = np.array([[1., 3.], [3., 1.], [5., 7.]])
train_y = np.array([190., 330., 660.]).reshape(3, 1)
print &quot;x=&quot;, train_x
print &quot;y=&quot;, train_y
※予測式で定義した形状(shape)に合わせること
※実用場面では，外部データを(ファイルやSQLなどから)
読みとって2次元配列に整形する。
190
330
660
1 3
3 1
5 7
train_x train_y
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-28-638.jpg?cb=1456578209" title="学習実行
2016-02-27 GDG京都 機械学習勉強会
28
# (2) セッションを準備し，変数を初期化
ses..." target="_blank">
        28.
      </a>
    学習実行
2016-02-27 GDG京都 機械学習勉強会
28
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
# (3) 最急勾配法でパラメータ更新 (100回更新する)
for i in range(100):
_, l, a_ = sess.run([train_step, loss, a], feed_dict={x: train_x, y_: train_y})
if (i + 1) % 10 == 0:
print &quot;step=%3d, a1=%6.2f, a2=%6.2f, loss=%.2f&quot; % (i + 1, a_[0], a_[1], l)
# (4) 学習結果を出力
est_a = sess.run(a, feed_dict={x: train_x, y_: train_y})
print &quot;Estimated: a1=%6.2f, a2=%6.2f&quot; % (est_a[0], est_a[1])
おまじない（初期化）
sess.run ( [出力, ...], feed_dict={入力リスト} )
sess.run を呼び出すことで、&quot;出力&quot;に指定した
データフローグラフが計算される
※ 学習を回すには、先ほど作成した最適化手段(train_step)を
出力値に指定して sess.run を呼び出す。
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-29-638.jpg?cb=1456578209" title="参考: 実行結果
2016-02-27 GDG京都 機械学習勉強会
29
うまく学習できると...
step= 10,..." target="_blank">
        29.
      </a>
    参考: 実行結果
2016-02-27 GDG京都 機械学習勉強会
29
うまく学習できると...
step= 10, a1= 70.35, a2= 46.23, loss=2189.06
step= 20, a1= 83.06, a2= 36.70, loss=771.90
step= 30, a1= 90.13, a2= 31.41, loss=334.34
step= 40, a1= 94.05, a2= 28.47, loss=199.24
step= 50, a1= 96.23, a2= 26.84, loss=157.52
step= 60, a1= 97.44, a2= 25.93, loss=144.64
step= 70, a1= 98.12, a2= 25.42, loss=140.67
step= 80, a1= 98.49, a2= 25.14, loss=139.44
step= 90, a1= 98.70, a2= 24.99, loss=139.06
step=100, a1= 98.81, a2= 24.90, loss=138.94
Estimated: a1= 98.81, a2= 24.90
順
調
に
低
下
推定されたパラメータの値
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-30-638.jpg?cb=1456578209" title="参考: 実行結果
2016-02-27 GDG京都 機械学習勉強会
30
学習率が大きすぎて学習失敗していると...
..." target="_blank">
        30.
      </a>
    参考: 実行結果
2016-02-27 GDG京都 機械学習勉強会
30
学習率が大きすぎて学習失敗していると...
step= 10, (中略), loss=72692296.00
step= 20, (中略), loss=54651076608.00
step= 30, (中略), loss=41087909494784.00
: :
順
調
に
増
加
学習率を小さくしてみる
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-31-638.jpg?cb=1456578209" title="4. 予測
2016-02-27 GDG京都 機械学習勉強会
31
# (1) 新しいデータを用意
new_x = n..." target="_blank">
        31.
      </a>
    4. 予測
2016-02-27 GDG京都 機械学習勉強会
31
# (1) 新しいデータを用意
new_x = np.array([2., 4.]).reshape(1, 2)
# (2) 学習結果をつかって，予測実施
new_y = sess.run(y, feed_dict={x: new_x})
print new_y
予測でも sess.run を用いる
feed_dictには新しい入力値を指定することに留意。(当然ですが...)
（参考）実行結果
[[ 297.22738647]]
円円 × 2 ＋ 円 × 4 ≒99 25 297
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-32-638.jpg?cb=1456578209" title="5. 後片付け
2016-02-27 GDG京都 機械学習勉強会
32
# セッションを閉じる
sess.close()
 " target="_blank">
        32.
      </a>
    5. 後片付け
2016-02-27 GDG京都 機械学習勉強会
32
# セッションを閉じる
sess.close()
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-33-638.jpg?cb=1456578209" title="TensorFlow の処理の流れ (再掲)
2016-02-27 GDG京都 機械学習勉強会
33
# coding..." target="_blank">
        33.
      </a>
    TensorFlow の処理の流れ (再掲)
2016-02-27 GDG京都 機械学習勉強会
33
# coding: utf-8
# 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
# 1. 予測式(モデル)を記述する
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=(None, 2), name=&quot;x&quot;)
y_ = tf.placeholder(tf.float32, shape=(None, 1), name=&quot;y&quot;)
# モデルパラメータ
a = tf.Variable(tf.zeros((2, 1)), name=&quot;a&quot;)
# モデル式
y = tf.matmul(x, a)
# 2. 学習に必要な関数を定義する
# 誤差関数(loss)
loss = tf.reduce_mean(tf.square(y_ - y))
# 最適化手段を選ぶ(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.02).minimize(loss)
# 3. 実際に学習処理を実行する
# (1) 訓練データを生成する
train_x = np.array([[1., 3.], [3., 1.], [5., 7.]])
train_y = np.array([190., 330., 660.]).reshape(3, 1)
print &quot;x=&quot;, train_x
print &quot;y=&quot;, train_y
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
# (3) 最急勾配法でパラメータ更新 (100回更新する)
for i in range(100):
_, l, a_ = sess.run([train_step, loss, a], feed_dict={x: train_x, y_: train_y})
if (i + 1) % 10 == 0:
print &quot;step=%3d, a1=%6.2f, a2=%6.2f, loss=%.2f&quot; % (i + 1, a_[0], a_[1], l)
# (4) 学習結果を出力
est_a = sess.run(a, feed_dict={x: train_x, y_: train_y})
print &quot;Estimated: a1=%6.2f, a2=%6.2f&quot; % (est_a[0], est_a[1])
# 4. 新しいデータに対して予測する
# (1) 新しいデータを用意
new_x = np.array([2., 4.]).reshape(1, 2)
# (2) 学習結果をつかって，予測実施
new_y = sess.run(y, feed_dict={x: new_x})
print new_y
# 5. 後片付け
# セッションを閉じる
sess.close()
0. ライブラリをロード
1. 予測式(モデル)を記述
2. 誤差関数と最適化手法を記述
3. 訓練データを作成(or読込)し
学習実行
4. 予測
5. 後片付け
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-34-638.jpg?cb=1456578209" title="(補足) 線形回帰の応用場面
 家賃
 家賃 = a * 駅からの距離 + b * 築年数 + c * 広さ
 ..." target="_blank">
        34.
      </a>
    (補足) 線形回帰の応用場面
 家賃
 家賃 = a * 駅からの距離 + b * 築年数 + c * 広さ
 明日の客数
 客数 = a * 気温 + b * 昨年同日の客数
などなど
2016-02-27 GDG京都 機械学習勉強会
34
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-35-638.jpg?cb=1456578209" title="つぎの問題です
2016-02-27 GDG京都 機械学習勉強会
35
たかしくんは八百屋へ財布を預かってお使いに
行..." target="_blank">
        35.
      </a>
    つぎの問題です
2016-02-27 GDG京都 機械学習勉強会
35
たかしくんは八百屋へ財布を預かってお使いに
行きました。しかしたかしくんはお金を
数えられません。
気まぐれおやじ曰く、
リンゴ２個＋ミカン３個、リンゴ０個＋ミカン１６個
なら買えるが、リンゴ３個＋ミカン１個、
リンゴ２個＋ミカン８個は買えないとのこと。
リンゴ１個＋ミカン１１個は買えますか？
例題２
識別問題
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-36-638.jpg?cb=1456578209" title="機械学習の流れ
2016-02-27 GDG京都 機械学習勉強会
36
学
習
予
測
①予測式(モデル)をつくる
②..." target="_blank">
        36.
      </a>
    機械学習の流れ
2016-02-27 GDG京都 機械学習勉強会
36
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-37-638.jpg?cb=1456578209" title="式で表そうとしてみる...
2016-02-27 GDG京都 機械学習勉強会
37
円 ×1＋ 円
円 ×2＋ 円 ×..." target="_blank">
        37.
      </a>
    式で表そうとしてみる...
2016-02-27 GDG京都 機械学習勉強会
37
円 ×1＋ 円
円 ×2＋ 円 × 3)
円 ×0＋ 円 ×16)
円 ×3＋ 円 × 1)
円
円
円
円 ×2＋ 円 × 8 )円
×11)円
-(
-(
-(
-(
-(
買える
買える
買えない
買えない
?
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-38-638.jpg?cb=1456578209" title="式で表そうとしてみる...
2016-02-27 GDG京都 機械学習勉強会
38
円 ×1＋ 円
円 ×2＋ 円 ×..." target="_blank">
        38.
      </a>
    式で表そうとしてみる...
2016-02-27 GDG京都 機械学習勉強会
38
円 ×1＋ 円
円 ×2＋ 円 × 3)
円 ×0＋ 円 ×16)
円 ×3＋ 円 × 1)
円
円
円
円 ×2＋ 円 × 8 )円
×11)円
-(
-(
-(
-(
-(
買える
買える
買えない
買えない
?
予想される残金 買える : 1
買えない : 0
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-39-638.jpg?cb=1456578209" title="シグモイド曲線
2016-02-27 GDG京都 機械学習勉強会
39
予想される残金
0
0.5
買える
買えない
..." target="_blank">
        39.
      </a>
    シグモイド曲線
2016-02-27 GDG京都 機械学習勉強会
39
予想される残金
0
0.5
買える
買えない
𝜎 𝑢 =
1
1 + exp(−𝑢)
ロジット
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-40-638.jpg?cb=1456578209" title="予測式(モデル)が作れた！
2016-02-27 GDG京都 機械学習勉強会
40
円 ×1＋ 円
円 ×2＋ 円 ×..." target="_blank">
        40.
      </a>
    予測式(モデル)が作れた！
2016-02-27 GDG京都 機械学習勉強会
40
円 ×1＋ 円
円 ×2＋ 円 × 3)) ＝ 1
円 ×0＋ 円 ×16)) ＝ 1
円 ×3＋ 円 × 1)) ＝ 0
円
円
円
円 ×2＋ 円 × 8 )) ＝ 0円
×11) ) ＝円
-(
-(
-(
-(
-( ?
σ(
σ(
σ(
σ(
σ(
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-41-638.jpg?cb=1456578209" title="ロジスティック回帰
2016-02-27 GDG京都 機械学習勉強会
41
入力層 出力層
※ロジスティック「回帰」で..." target="_blank">
        41.
      </a>
    ロジスティック回帰
2016-02-27 GDG京都 機械学習勉強会
41
入力層 出力層
※ロジスティック「回帰」ですが，識別(分類)の手法です。
𝑦 = 𝜎(𝑏 + 𝑎1 𝑥1 + 𝑎2 𝑥2)
𝑦
𝑥1
𝑥2
× 𝑎1
× 𝑎2
リンゴの数 
ミカンの数 
買えるか
リンゴ
の単価
↓
↑
ミカン
の単価
𝑏
財布の中身
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-42-638.jpg?cb=1456578209" title="TensorFlow での実装
2016-02-27 GDG京都 機械学習勉強会
42
# coding: utf-8..." target="_blank">
        42.
      </a>
    TensorFlow での実装
2016-02-27 GDG京都 機械学習勉強会
42
# coding: utf-8
# ### 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
# ## TensorFlow でロジスティック回帰する
# 1. 学習したいモデルを記述する
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=(None, 2), name=&quot;x&quot;)
y_ = tf.placeholder(tf.float32, shape=(None, 1), name=&quot;y&quot;)
# モデルパラメータ
a = tf.Variable(-10 * tf.ones((2, 1)), name=&quot;a&quot;)
b = tf.Variable(200., name=&quot;b&quot;)
# モデル式
u = tf.matmul(x, a) + b
y = tf.sigmoid(u)
# 2. 学習やテストに必要な関数を定義する
# 誤差関数(loss)
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(u, y_))
# 最適化手段(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
# 3. 実際に学習処理を実行する
# (1) 訓練データを生成する
train_x = np.array([[2., 3.], [0., 16.], [3., 1.], [2., 8.]])
train_y = np.array([1., 1., 0., 0.]).reshape(4, 1)
print &quot;x=&quot;, train_x
print &quot;y=&quot;, train_y
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
# (3) 最急勾配法でパラメータ更新 (1000回更新する)
for i in range(1000):
_, l, a_, b_ = sess.run([train_step, loss, a, b], feed_dict={x: train_x, y_: train_y})
if (i + 1) % 100 == 0:
print &quot;step=%3d, a1=%6.2f, a2=%6.2f, b=%6.2f, loss=%.2f&quot; % (i + 1, a_[0], a_[1], b_, l)
# (4) 学習結果を出力
est_a, est_b = sess.run([a, b], feed_dict={x: train_x, y_: train_y})
print &quot;Estimated: a1=%6.2f, a2=%6.2f, b=%6.2f&quot; % (est_a[0], est_a[1], est_b)
# 4. 新しいデータに対して予測する
# (1) 新しいデータを用意
new_x = np.array([1., 11.]).reshape(1, 2)
# (2) 学習結果をつかって，予測実施
new_y = sess.run(y, feed_dict={x: new_x})
print new_y
# 5. 後片付け
# セッションを閉じる
sess.close()
0. ライブラリをロード
1. 予測式(モデル)を記述
2. 誤差関数と最適化手法を記述
3. 訓練データを作成(or読込)し
学習実行
4. 予測
5. 後片付け
※線形回帰の実装と，ほとんど同一
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-43-638.jpg?cb=1456578209" title="1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
43
# 1. 学習したいモデルを記述..." target="_blank">
        43.
      </a>
    1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
43
# 1. 学習したいモデルを記述する
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, shape=(None, 2), name=&quot;x&quot;)
y_ = tf.placeholder(tf.float32, shape=(None, 1), name=&quot;y&quot;)
# モデルパラメータ
a = tf.Variable(-10 * tf.ones((2, 1)), name=&quot;a&quot;)
b = tf.Variable(200., name=&quot;b&quot;)
# モデル式
u = tf.matmul(x, a) + b
y = tf.sigmoid(u)
𝑦 = 𝜎(𝑏 + 𝑎1 𝑥1 + 𝑎2 𝑥2)
ロジット: u
今回の予測式に合わせて
モデルとパラメータを修正
追加
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-44-638.jpg?cb=1456578209" title="2. 誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
44
# 誤差関数(loss)
l..." target="_blank">
        44.
      </a>
    2. 誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
44
# 誤差関数(loss)
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(u, y_))
# 最適化手段(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
・誤差関数を変更
識別(分類)問題  クロスエントロピー
−𝑦𝑖 log 𝑦𝑖 − 1 − 𝑦𝑖 log(1 − 𝑦𝑖)
𝑁
𝑖
正解
予測値: 𝑦𝑖
正解
予測値: 𝑦𝑖
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-45-638.jpg?cb=1456578209" title="3. 学習実行
2016-02-27 GDG京都 機械学習勉強会
45
# (2) セッションを準備し，変数を初期化
..." target="_blank">
        45.
      </a>
    3. 学習実行
2016-02-27 GDG京都 機械学習勉強会
45
# (2) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
# (3) 最急勾配法でパラメータ更新 (1000回更新する)
for i in range(1000):
_, l, a_, b_ = sess.run([train_step, loss, a, b], feed_dict={x: train_x, y_: train_y})
if (i + 1) % 100 == 0:
print &quot;step=%3d, a1=%6.2f, a2=%6.2f, b=%6.2f, loss=%.2f&quot; % (i + 1, a_[0], a_[1], b_, l)
# (4) 学習結果を出力
est_a, est_b = sess.run([a, b], feed_dict={x: train_x, y_: train_y})
print &quot;Estimated: a1=%6.2f, a2=%6.2f, b=%6.2f&quot; % (est_a[0], est_a[1], est_b)
線形回帰とほぼ同じ！
【変更箇所】
・パラメータ b の出力を追加
・更新回数を 100回  1000回 に
※更新回数は，対象問題やデータ，初期値，モデルなどでまちまちです。
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-46-638.jpg?cb=1456578209" title="学習結果
2016-02-27 GDG京都 機械学習勉強会
46
step=100, a1=-22.50, a2=-1..." target="_blank">
        46.
      </a>
    学習結果
2016-02-27 GDG京都 機械学習勉強会
46
step=100, a1=-22.50, a2=-12.28, b=196.26, loss=42.75
step=200, a1=-35.00, a2=-12.06, b=192.68, loss=25.84
step=300, a1=-47.36, a2=-11.78, b=189.14, loss=9.24
step=400, a1=-55.13, a2=-11.51, b=186.75, loss=2.54
step=500, a1=-58.92, a2=-11.29, b=185.58, loss=0.02
step=600, a1=-59.26, a2=-11.23, b=185.47, loss=0.01
step=700, a1=-59.43, a2=-11.19, b=185.43, loss=0.00
step=800, a1=-59.53, a2=-11.17, b=185.39, loss=0.00
step=900, a1=-59.62, a2=-11.15, b=185.37, loss=0.00
step=1000, a1=-59.68, a2=-11.14, b=185.35, loss=0.00
Estimated: a1=-59.68, a2=-11.14, b=185.35
順
調
に
低
下
推定されたパラメータの値
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-47-638.jpg?cb=1456578209" title="予測結果
2016-02-27 GDG京都 機械学習勉強会
47
円 ×1＋ 円 ×11) ) ＝円 -( 0.96σ..." target="_blank">
        47.
      </a>
    予測結果
2016-02-27 GDG京都 機械学習勉強会
47
円 ×1＋ 円 ×11) ) ＝円 -( 0.96σ( 185 60 11
リンゴ１個＋ミカン１１個は，おそらく買えそう
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-48-638.jpg?cb=1456578209" title="最後にもう少し複雑な問題
2016-02-27 GDG京都 機械学習勉強会
48
手書き文字認識
MNISTデータ
0..." target="_blank">
        48.
      </a>
    最後にもう少し複雑な問題
2016-02-27 GDG京都 機械学習勉強会
48
手書き文字認識
MNISTデータ
0～9の手書き文字画像を7万枚集めたデータセット
1枚の画像は 28 x 28 ピクセル
7 3 4 6
多クラス識別(分類)問題
入力：
出力：
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-49-638.jpg?cb=1456578209" title="機械学習の流れ
2016-02-27 GDG京都 機械学習勉強会
49
学
習
予
測
①予測式(モデル)をつくる
②..." target="_blank">
        49.
      </a>
    機械学習の流れ
2016-02-27 GDG京都 機械学習勉強会
49
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-50-638.jpg?cb=1456578209" title="one-hot ベクトル (one-of-K表現)
2016-02-27 GDG京都 機械学習勉強会
50
7
3
4..." target="_blank">
        50.
      </a>
    one-hot ベクトル (one-of-K表現)
2016-02-27 GDG京都 機械学習勉強会
50
7
3
4
6
出力ラベル(1次元) 出力ラベル(10次元)
０ 0 0 0 0 0 0 1 0 0
0 1 2 3 4 5 6 7 8 9
０ 0 0 1 0 0 0 0 0 0
０ 0 0 0 1 0 0 0 0 0
０ 0 0 0 0 0 1 0 0 0
1つの多クラス識別問題が
10個の2クラス識別問題へ分解
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-51-638.jpg?cb=1456578209" title="案1: ロジスティック回帰を拡張
2016-02-27 GDG京都 機械学習勉強会
51
入力層
(768ノード)
出..." target="_blank">
        51.
      </a>
    案1: ロジスティック回帰を拡張
2016-02-27 GDG京都 機械学習勉強会
51
入力層
(768ノード)
出力層
(10ノード)
𝑦0
𝑦1
𝑦9
ソフトマックス(Softmax) 𝑦 = 𝑠𝑜𝑓𝑡𝑚𝑎𝑥(𝑥𝑊 + 𝑏)
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-52-638.jpg?cb=1456578209" title="案1: ロジスティック回帰を拡張
2016-02-27 GDG京都 機械学習勉強会
52
入力層
(768ノード)
出..." target="_blank">
        52.
      </a>
    案1: ロジスティック回帰を拡張
2016-02-27 GDG京都 機械学習勉強会
52
入力層
(768ノード)
出力層
(10ノード)
𝑦0
𝑦1
𝑦9
ソフトマックス(Softmax)
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-53-638.jpg?cb=1456578209" title="案2: さらに中間層を追加
2016-02-27 GDG京都 機械学習勉強会
53
入力層
(768ノード)
出力層
..." target="_blank">
        53.
      </a>
    案2: さらに中間層を追加
2016-02-27 GDG京都 機械学習勉強会
53
入力層
(768ノード)
出力層
(10ノード)
𝑦0
𝑦1
𝑦9
ニューラルネット（パーセプトロン）
隠れ層
(□ノード)
中間層(隠れ層)を追加すると，より複雑なパターンも見分けられる
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-54-638.jpg?cb=1456578209" title="TensorFlow での実装
2016-02-27 GDG京都 機械学習勉強会
54
# 0. 必要なモジュールを読..." target="_blank">
        54.
      </a>
    TensorFlow での実装
2016-02-27 GDG京都 機械学習勉強会
54
# 0. 必要なモジュールを読み込む
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
# MNISTデータの取得
mnist = input_data.read_data_sets(&quot;MNIST_data/&quot;, one_hot=True)
# 1. 学習したいモデルを記述する
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
# モデルパラメータ(入力層:784ノード, 隠れ層:100ノード, 出力層:10ノード)
W1 = tf.Variable(tf.truncated_normal([784, 100]))
b1 = tf.Variable(tf.zeros([100]))
W2 = tf.Variable(tf.truncated_normal([100, 10]))
b2 = tf.Variable(tf.zeros([10]))
# モデル式
h = tf.sigmoid(tf.matmul(x, W1) + b1) # 入力層-&gt;隠れ層
u = tf.matmul(h, W2) + b2 # 隠れ層-&gt;出力層 (ロジット)
y = tf.nn.softmax(u) # 隠れ層-&gt;出力層 (ソフトマックス後)
# 2. 学習やテストに必要な関数を定義する
# 誤差関数(loss)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(u, y_))
# 最適化手段(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
# 正答率(学習には用いない)
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
# 3. 実際に学習処理を実行する
# (1) セッションを準備し，変数を初期化
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
# (2) バッチ型確率的勾配降下法でパラメータ更新
for i in range(10000):
# 訓練データから100サンプルをランダムに取得
batch_xs, batch_ys = mnist.train.next_batch(100)
# 学習
_, l = sess.run([train_step, loss], feed_dict={x: batch_xs, y_: batch_ys})
if (i + 1) % 1000 == 0:
print &quot;step=%3d, loss=%.2f&quot; % (i + 1, l)
# 4. テスト用データに対して予測し，性能を確認
# (1) テスト用データを1000サンプル取得
new_x = mnist.test.images[0:1000]
new_y_ = mnist.test.labels[0:1000]
# (2) 予測と性能評価
accuracy, new_y = sess.run([acc, y], feed_dict={x:new_x , y_:new_y_ })
print &quot;Accuracy (for test data): %6.2f%%&quot; % accuracy
print &quot;True Label:&quot;, np.argmax(new_y_[0:15,], 1)
print &quot;Est Label:&quot;, np.argmax(new_y[0:15, ], 1)
# 5. 後片付け
# セッションを閉じる
sess.close()
0. ライブラリをロード
1. 予測式(モデル)を記述
2. 誤差関数と最適化手法を記述
3. 訓練データを作成(or読込)し
学習実行
4. 予測
5. 後片付け
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-55-638.jpg?cb=1456578209" title="1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
55
# 入力変数と出力変数のプレース..." target="_blank">
        55.
      </a>
    1. 予測式(モデル)を記述
2016-02-27 GDG京都 機械学習勉強会
55
# 入力変数と出力変数のプレースホルダを生成
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
# モデルパラメータ(入力層:784ノード, 隠れ層:100ノード, 出力層:10ノード)
W1 = tf.Variable(tf.truncated_normal([784, 100]))
b1 = tf.Variable(tf.zeros([100]))
W2 = tf.Variable(tf.truncated_normal([100, 10]))
b2 = tf.Variable(tf.zeros([10]))
# モデル式
h = tf.sigmoid(tf.matmul(x, W1) + b1) # 入力層-&gt;隠れ層
u = tf.matmul(h, W2) + b2 # 隠れ層-&gt;出力層 (ロジット)
y = tf.nn.softmax(u) # 隠れ層-&gt;出力層 (ソフトマックス後)
ｘ
𝑥𝑊1 + 𝑏1
h
𝑠𝑖𝑔𝑚𝑜𝑖𝑑()
隠れ層
u y
ℎ𝑊2 + 𝑏2 𝑠𝑜𝑓𝑡𝑚𝑎𝑥()
出力層入力層
784
10
予測式(モデル, ネットワーク)を実装
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-56-638.jpg?cb=1456578209" title="2. 誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
56
# 誤差関数(loss)
l..." target="_blank">
        56.
      </a>
    2. 誤差関数と最適化手法を記述
2016-02-27 GDG京都 機械学習勉強会
56
# 誤差関数(loss)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(u, y_))
# 最適化手段(最急降下法)
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
# 正答率(学習には用いない)
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
・誤差関数を変更
多クラス識別(分類)問題
 多クラス用クロスエントロピー
(softmax_cross_entropy_with_logits)
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-57-638.jpg?cb=1456578209" title="3. 学習実行
2016-02-27 GDG京都 機械学習勉強会
57
# MNISTデータの取得
mnist = i..." target="_blank">
        57.
      </a>
    3. 学習実行
2016-02-27 GDG京都 機械学習勉強会
57
# MNISTデータの取得
mnist = input_data.read_data_sets(&quot;MNIST_data/&quot;, one_hot=True)
# (2) バッチ型確率的勾配降下法でパラメータ更新
for i in range(10000):
# 訓練データから100サンプルをランダムに取得
batch_xs, batch_ys = mnist.train.next_batch(100)
# 学習
_, l = sess.run([train_step, loss], feed_dict={x: batch_xs, y_: batch_ys})
if (i + 1) % 1000 == 0:
print &quot;step=%3d, loss=%.2f&quot; % (i + 1, l)
大規模なデータで学習する時の工夫
・各ステップで，100個の訓練データを
ランダムに取り出して学習 (確率的勾配降下法)
 速く学習が進む
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-58-638.jpg?cb=1456578209" title="参考: 学習結果
2016-02-27 GDG京都 機械学習勉強会
58
step=1000, loss=2.09
s..." target="_blank">
        58.
      </a>
    参考: 学習結果
2016-02-27 GDG京都 機械学習勉強会
58
step=1000, loss=2.09
step=2000, loss=1.45
step=3000, loss=1.38
step=4000, loss=0.77
step=5000, loss=0.63
step=6000, loss=0.52
step=7000, loss=0.78
step=8000, loss=0.73
step=9000, loss=0.55
step=10000, loss=0.58
ゆ
ら
ぎ
な
が
ら
低
下
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-59-638.jpg?cb=1456578209" title="4. 予測
2016-02-27 GDG京都 機械学習勉強会
59
# (1) テスト用データを1000サンプル取得
..." target="_blank">
        59.
      </a>
    4. 予測
2016-02-27 GDG京都 機械学習勉強会
59
# (1) テスト用データを1000サンプル取得
new_x = mnist.test.images[0:1000]
new_y_ = mnist.test.labels[0:1000]
# (2) 予測と性能評価
accuracy, new_y = sess.run([acc, y], feed_dict={x:new_x , y_:new_y_ })
print &quot;Accuracy (for test data): %6.2f%%&quot; % accuracy * 100
print &quot;True Label:&quot;, np.argmax(new_y_[0:15,], 1)
print &quot;Est Label:&quot;, np.argmax(new_y[0:15, ], 1)
ここはこれまでと同様
【実行結果例】
Accuracy(test data): 80.0%
True Label: [7 2 1 0 4 1 4 9 5 9 0 6 9 0 1]
Est Label: [7 2 1 0 9 1 4 9 2 9 0 6 9 0 1]
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-60-638.jpg?cb=1456578209" title="まとめ（機械学習の流れ）
2016-02-27 GDG京都 機械学習勉強会
60
学
習
予
測
①予測式(モデル)を..." target="_blank">
        60.
      </a>
    まとめ（機械学習の流れ）
2016-02-27 GDG京都 機械学習勉強会
60
学
習
予
測
①予測式(モデル)をつくる
②誤差を計算する
③誤差を最小にする点を探す(最適化)
予測式&amp;学習済パラメタで予測
 
  </li>
  <li>
      <a href="http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-61-638.jpg?cb=1456578209" title="参考書籍
2016-02-27 GDG京都 機械学習勉強会
61
ディープラーニングの基礎的な理論と
TensorFl..." target="_blank">
        61.
      </a>
    参考書籍
2016-02-27 GDG京都 機械学習勉強会
61
ディープラーニングの基礎的な理論と
TensorFlowなどのサンプルコードあり
ディープラーニングの
理論理解を深めたいかたはこちら。
 
  </li>

              </ol>
            </div>
      </div>
    </div>

    <aside id="side-panel" class="small-12 large-4 columns j-related-more-tab">
          <div class="mediumRectangleAd">
    <div id="topRightAd" class="j-top-right-div top-right-fillin">
      <script type="text/javascript"> slideshare_object._adQueue.push({ tile : 5, zone : 'slideview', dart_code: '', width: 300, height: 250, appendTo: 'topRightAd'}); </script>
    </div>
  </div>

<dl class="tabs related-tabs small" data-tab>
    <dd class="active">
      <a href="#related-tab-content" data-ga-cat="bigfoot_slideview" data-ga-action="relatedslideshows_tab">
        Recommended
      </a>
    </dd>
</dl>

<div class="tabs-content">
    <ul id="related-tab-content" class="content active no-bullet notranslate">

        <li class="j-related-item">
  <a data-ssid="42758507"
    data-algo-id=""
    data-source-name="MORE_FROM_USER"
    data-source-model=""
    data-urn-type="Slideshow"
    data-score=""
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="BigQuery + Fluentd"
    href="/ToruUenoyama/20141213-big-queryandfluentd">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="BigQuery + Fluentd"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/20141213bigqueryandfluentd-141216082129-conversion-gate01-thumbnail-2.jpg?cb=1418718246" />
    </div>

    <div class="related-content">
      <div class="title">
        BigQuery + Fluentd
      </div>

        <div class="author">Toru UENOYAMA</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="12270323"
    data-algo-id=""
    data-source-name="MORE_FROM_USER"
    data-source-model=""
    data-urn-type="Slideshow"
    data-score=""
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="Reading android.hardware.Camera"
    href="/ToruUenoyama/reading-androidhardwarecamera">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="Reading android.hardware.Camera"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/androidhardwarecamera20120404-120403101126-phpapp01-thumbnail-2.jpg?cb=1333448206" />
    </div>

    <div class="related-content">
      <div class="title">
        Reading android.hardware.Camera
      </div>

        <div class="author">Toru UENOYAMA</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="11634729"
    data-algo-id=""
    data-source-name="MORE_FROM_USER"
    data-source-model=""
    data-urn-type="Slideshow"
    data-score=""
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="OpenCVをAndroidで動かしてみた"
    href="/ToruUenoyama/opencvandroid">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="OpenCVをAndroidで動かしてみた"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/2011-12-03opencvandroid-120217073744-phpapp02-thumbnail-2.jpg?cb=1329464494" />
    </div>

    <div class="related-content">
      <div class="title">
        OpenCVをAndroidで動かしてみた
      </div>

        <div class="author">Toru UENOYAMA</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58750805"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="Introduction to Information Architecture"
    href="/mikecrabb/introduction-to-information-architecture-58750805">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="Introduction to Information Architecture"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/introductiontoinformationarchitecture-160226103935-thumbnail-2.jpg?cb=1456521962" />
    </div>

    <div class="related-content">
      <div class="title">
        Introduction to Information Architecture
      </div>

        <div class="author">Mike Crabb</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58736642"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="The Link Between Alcohol and Breast Cancer"
    href="/omerhameed/the-link-between-alcohol-and-breast-cancer">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="The Link Between Alcohol and Breast Cancer"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/risk-of-breast-cancer-160226020800-thumbnail-2.jpg?cb=1456521687" />
    </div>

    <div class="related-content">
      <div class="title">
        The Link Between Alcohol and Breast Cancer
      </div>

        <div class="author">Dr.  Omer Hameed</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58754465"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="The Death of Polling?"
    href="/IpsosMORI/the-death-of-polling">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="The Death of Polling?"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/thedeathofpolling-160226121628-thumbnail-2.jpg?cb=1456521635" />
    </div>

    <div class="related-content">
      <div class="title">
        The Death of Polling?
      </div>

        <div class="author">Ipsos MORI</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58756773"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="Publishing Production: From the Desktop to the Cloud"
    href="/DeantaGlobal/publishing-production-from-the-desktop-to-the-cloud">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="Publishing Production: From the Desktop to the Cloud"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/desktoptothecloud-160226132022-thumbnail-2.jpg?cb=1456521576" />
    </div>

    <div class="related-content">
      <div class="title">
        Publishing Production: From the Desktop to the Cloud
      </div>

        <div class="author">Deanta </div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58772956"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="Mobile App Prototyping for Designers"
    href="/jonoYoung/1280x800geekendprotypingfordesigners">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="Mobile App Prototyping for Designers"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/0c780e37-2098-4e84-971b-160c56b9f3ec-160226200020-thumbnail-2.jpg?cb=1456521500" />
    </div>

    <div class="related-content">
      <div class="title">
        Mobile App Prototyping for Designers
      </div>

        <div class="author">Jono Young</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58772277"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="How Email and Great Content Can Fuel Your Buyer&#x27;s Journey"
    href="/visually/how-email-and-great-content-can-fuel-your-buyers-journey">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="How Email and Great Content Can Fuel Your Buyer&#x27;s Journey"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/visuallyemmadeck-edit-3-160226193745-thumbnail-2.jpg?cb=1456521484" />
    </div>

    <div class="related-content">
      <div class="title">
        How Email and Great Content Can Fuel Your Buyer&#x27;s Journey
      </div>

        <div class="author">Visually</div>

    </div>
  </a>
</li>

        <li class="j-related-item">
  <a data-ssid="58771713"
    data-algo-id="-2"
    data-source-name="FEATURED"
    data-source-model="-2"
    data-urn-type="Slideshow"
    data-score="0"
    class="j-related-impression slideview_related_item j-recommendation-tracking"
    title="This Was a Huge Week for Energy Stocks"
    href="/TheMotleyFool/this-was-a-huge-week-for-energy-stocks">

    <div class="related-thumbnail">
      <img class="j-thumbnail j-lazy-thumb"
        alt="This Was a Huge Week for Energy Stocks"
        src="//public.slidesharecdn.com/b/images/thumbnail.png"
        data-original="//cdn.slidesharecdn.com/ss_thumbnails/energystocksrocketedhigherthisweek-160226192053-thumbnail-2.jpg?cb=1456521364" />
    </div>

    <div class="related-content">
      <div class="title">
        This Was a Huge Week for Energy Stocks
      </div>

        <div class="author">The Motley Fool</div>

    </div>
  </a>
</li>

    </ul>

</div>

    </aside>
  </div>
</div>


<div id="first-clip-tour" class="hide">
</div>

<div id="clip-intention-tour" class="hide">
</div>


  <div class="footer-leaderboard-ad" style="text-align:center; padding-top: 15px;">
    <div class="transcriptAd">
      <div id="transcriptAdHolder"><script type="text/javascript">slideshare_object._adQueue.push({ tile : 4, zone : 'slideview', dart_code: '%3Bcat%3Dtechnology', width: 728, height: 90, appendTo: 'transcriptAdHolder', lazyLoad: true});</script></div>
    </div>
  </div>


      

      

        <footer>
            <div class="row">
              <div class="columns">
                <div id="smt-lang-selector"></div>
                <ul class="j-languages-selector language-links text-center">
                  <li class="smt-item j-www">
                    <a class="smt-link"
                      href="http://www.slideshare.net/ToruUenoyama/tensorflow-gdg?smtNoRedir=1"
                      title="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会) - English" lang="en" hreflang="en">English
                    </a>
                  </li>
                  <li class="smt-item j-es">
                    <a class="smt-link"
                      href="http://es.slideshare.net/ToruUenoyama/tensorflow-gdg"
                      title="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会) - Espanol" lang="es" hreflang="es">Espanol
                    </a>
                  </li>
                  <li class="smt-item j-pt">
                    <a class="smt-link"
                      href="http://pt.slideshare.net/ToruUenoyama/tensorflow-gdg"
                      title="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会) - Portugues" lang="pt" hreflang="pt">Portugues
                    </a>
                  </li>
                  <li class="smt-item j-fr">
                    <a class="smt-link"
                      href="http://fr.slideshare.net/ToruUenoyama/tensorflow-gdg"
                      title="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会) - Fran&ccedil;ais" lang="fr" hreflang="fr">Fran&ccedil;ais
                    </a>
                  </li>
                  <li class="smt-item j-de">
                    <a class="smt-link"
                      href="http://de.slideshare.net/ToruUenoyama/tensorflow-gdg"
                      title="TensorFlow を使った機械学習ことはじめ (GDG京都 機械学習勉強会) - Deutsche" lang="de" hreflang="de">Deutsche
                    </a>
                  </li>
                </ul>
              </div>
            </div>

          <div class="row">
            <div class="columns">
              <ul class="main-links text-center">
                <li><a href="/about">About</a></li>
                <li class="hidden-for-small"><a href="/developers">Dev & API</a></li>
                <li><a href="http://blog.slideshare.net/">Blog</a></li>
                <li><a href="/terms">Terms</a></li>
                <li><a href="/privacy">Privacy</a></li>
                <li><a href="http://www.linkedin.com/legal/copyright-policy">Copyright</a></li>
                <li class="hidden-for-small"><a href="https://www.linkedin.com/help/slideshare">Support</a></li>
              </ul>
            </div>
          </div>


          <div class="row">
            <div class="columns">
              <ul class="social-links text-center">
                <li>
                  <a title="Follow us on LinkedIn" href="http://www.linkedin.com/company/slideshare" class="fa fa-linkedin-square fa-lg" rel="nofollow" target="_blank"></a>
                </li>
                <li>
                  <a title="Follow us on Facebook" href="http://www.facebook.com/slideshare" class="fa fa-facebook-square fa-lg" rel="nofollow" target="_blank"></a>
                </li>
                <li>
                  <a title="Follow us on Twitter" href="http://twitter.com/SlideShare" class="fa fa-twitter-square fa-lg" rel="nofollow" target="_blank"></a>
                </li>
                <li>
                  <a title="Follow us on Google+" href="http://www.google.com/+SlideShare" class="fa fa-google-plus-square fa-lg" rel="nofollow" target="_blank"></a>
                </li>
                <li>
                  <a href="http://www.slideshare.net/rss/latest" class="fa fa-rss-square fa-lg"></a>
                </li>
              </ul>
            </div>
          </div>

          <div class="row">
            <div class="columns">
              <p class="copyright text-center">LinkedIn Corporation &copy; 2016</p>
              <p></p>
            </div>
          </div>
        </footer>
    </div>

    <div id="alert-modal" class="reveal-modal" data-reveal="">
      <p></p>
      <a class="close-reveal-modal">×</a>
    </div>

    <div class="modal_popup_container">
          
<div id="clipboard-share-modal" class="reveal-modal small mobile-hide" aria-hidden="true" aria-labelledby="modal-title" role="dialog" data-reveal data-ga-track-category="" data-ga-track-action="" translate="no">
  <div class="j-modal-popup modal-popup">
    <div id="modal-content" class="j-modal-content">
      <h4 class="j-modal-title modal-title notranslate">Share Clipboard</h4>
      <hr />
      <a class="close-reveal-modal" href="#" aria-label="Close">&times;</a>
      <div class="section share-email">
        <form class="j-share-email-form">
          <h5>Email</h5>
          <input class="j-share-email-to j-email-clear notranslate" name="recipients" placeholder="Enter email addresses" title="Enter email addresses" type="text">
          <div class="clearfix">
            <input data-ga="name" class="j-share-email-name j-email-clear notranslate" name="name" type="text" placeholder="From" title="From">
            <textarea class="j-share-email-msg j-email-clear notranslate share-message-textarea" name="message" placeholder="Add a message" title="Add a message"></textarea>
            <div class="j-email-flash email-flash"></div>
            <input id="share-email-send"class="button btn btn-inverse email-send-button notranslate" title="Send" type="submit" value="Send">
          </div>
        </form>
        <div id="email-sent" class="j-email-sent sent-section">
          <span class="success-text notranslate">Email sent successfully..</span>
        </div>
      </div>
      <div class="row">
        <ul class="j-share-social-list share-social-list" data-canonical-url="">
          <li class="facebook" data-network="facebook">
            <a class="share-link"
              rel="nofollow"
              title="Share on Facebook">Facebook</a>
          </li>
          <li class="twitter" data-network="twitter">
            <a class="share-link"
              rel="nofollow"
              title="Tweet on Twitter">Twitter</a>
          </li>
          <li class="linkedin" data-network="linkedin">
            <a class="share-link"
              rel="nofollow"
              title="Share on LinkedIn">LinkedIn</a>
          </li>
          <li class="googleplus" data-network="googleplus">
            <div class="social-hover">
              <a class="share-link"
                rel="nofollow"
                title="Share on Google+">Google+</a>
            </div>
          </li>
        </ul>
      </div>
      <div class="row">
        <div class="share-link-container">
          <label for="share-link-url">Link</label>
          <input id="share-link-url" class="j-share-link-url" type="text" data-ga="link"></input>
        </div>
      </div>
    </div>
  </div>
</div>

    <div id="top-clipboards-modal"
     class="reveal-modal xlarge top-clipboards-modal"
     data-reveal
     aria-labelledby="modal-title"
     aria-hidden="true"
     role="dialog">
  <h4 class="modal-title">Public clipboards featuring this slide</h4>
  <hr />
  <button class="close-reveal-modal button-lrg" aria-label="Close">&times;</button>
  <div class="loading text-center">
    <svg><use data-size="small" xlink:href="#loader"></use></svg>
  </div>
  <div class="empty">
    No public clipboards found for this slide
  </div>

  <div class="clipboards row">
    <ul class="small-block-grid-1 medium-block-grid-2 large-block-grid-3"></ul>
  </div>
</div>

    
<div id="download-interstitial-modal" class="download-interstitial reveal-modal medium" aria-hidden="false" role="dialog" data-reveal>
  <div class="modal-content-container">
    <a class="close-reveal-modal" href="#" aria-label="Close">&times;</a>
    <div class="modal-content">
      <div class="modal-inner-content">
        <div class="row">
          <h3 class="text-center">Save the most important slides with Clipping</h3>
        </div>
        <div class="row">
          <!-- make only 8/12 columns and centered -->
          <div class="medium-10 small-centered columns">
            <p class="text-center">Clipping is a handy way to collect and organize the most important slides from a presentation. You can keep your great finds in clipboards organized around topics.</p>
            <button class="art-deco small primary start-clipping">Start clipping</button>
            <button class="art-deco small tertiary button continue-download" data-reveal-id="login_modal" href="/login?from_source=%2FToruUenoyama%2Ftensorflow-gdg%3Ffrom_action%3Dsave&amp;from=download&amp;layout=foundation">No thanks. Continue to download.</button>
          </div>
        </div>
      </div>
    </div>
  </div>
</div>

    <div id="select-clipboard-modal" class="reveal-modal medium" data-reveal aria-labelledby="modal-title" aria-hidden="true" role="dialog" translate="no">
  <div class="modal-header-container">
    <h4 class="modal-title">Select another clipboard</h4>
    <hr />
    <a class="close-reveal-modal button-lrg" href="#" aria-label="Close">&times;</a>
  </div>

  <div class="loading">
    <svg><use data-size="small" xlink:href="#loader"></use></svg>
  </div>

  <div class="modal-content">

    <div class="default-clipboard-panel radius">
      <p>Looks like you’ve clipped this slide to <strong class="default-clipboard-title"></strong> already.</p>
    </div>

    <div class="clipboard-search-input-container">
      <i class="fa fa-search"></i>
      <input class="clipboard-search-input" type="text" placeholder="Search for a clipboard">
    </div>

    <div class="clipboard-list-container">
      <ul class="existing-clipboard-list"></ul>
      <ul class="clipboard-list"></ul>
      <div class="clipboard-create-new">
        <div class="clipboard-item-thumb">
          <i class="fa fa-plus"></i>
        </div>
        <div class="clipboard-item-title">Create a clipboard</div>
      </div>
    </div>
  </div>
</div>

    <div id="clipboard-create-modal" class="reveal-modal medium" data-reveal aria-labelledby="modal-title" aria-hidden="true" role="dialog" translate="no">
    <div class="first-clip-message hide">
      <h3>You just clipped your first slide!</h3>
      <span>
        Clipping is a handy way to collect important slides you want to go back to later. Now customize the name of a clipboard to store your clips.
      </span>
    </div>
    <h4 class="modal-title"></h4>
    <hr />
    <form data-abide>
    <div class="row">
      <div class="j-title-wrapper input-wrapper notranslate">
        <label>Name*
          <input type="text" class="j-title-input" placeholder="Best of Slides"/>
          <small class="j-title-error error hide"></small>
        </label>
      </div>
    </div>

    <div class="row">
      <div class="j-description-wrapper input-wrapper notranslate">
        <label>Description
          <input type="text" class="j-description-input" placeholder="Add a brief description so others know what your Clipboard is about."/>
          <small class="j-description-error error hide"></small>
        </label>
      </div>
    </div>

    <div class="row">
      <label>Visibility
        <div class="switch small notranslate">
          <small id="privacy-switch-description">Others can see my Clipboard</small>
          <input id="privacy-switch" class="j-privacy-input" type="checkbox" />
          <label for="privacy-switch"></label>
        </div>
      </label>
    </div>

        <div class="row">
          <div class="pull-right">
            <button class="j-cancel-button button-ghost small notranslate" type="button">Cancel</button>
            <button class="j-submit-button button-linkedin small notranslate" type="submit">Save</button>
          </div>
        </div>
    </form>
    <input type="hidden" class="j-i18n-edit-clipboard-modal" value='{&quot;formTitleEdit&quot;:&quot;Edit Clipboard&quot;,&quot;privacyDisabled&quot;:&quot;Others can see my Clipboard&quot;,&quot;privacyEnabled&quot;:&quot;Clipboard is visible only to me.&quot;,&quot;formTitleCreate&quot;:&quot;Create a clipboard&quot;}'
    />

</div>


    </div>

    <script src="http://public.slidesharecdn.com/b/ss_foundation/combined_foundation_base.js?e3dade9232" type="text/javascript"></script>

    <script>
      $(window).load(function() {
        loadCSS("//public.slidesharecdn.com/b/ss_foundation/stylesheets/app.css?7c4cd8d525165ef63899ad9317a2180252ea58a4");
      });
    </script>

      <meta content="{&quot;jsplayer_toolbar_clips.clip_count_info&quot;:&quot;{:count,choice,0#Be the first to clip this slide|singular#person clipped this slide|plural#people clipped this slide}&quot;}" name="ss-i18n-translations" />

  <meta content="{&quot;lynda.domain&quot;:&quot;lynda&quot;,&quot;selected&quot;:&quot;Selected&quot;,&quot;clipping.toast.change_clipboard&quot;:&quot;Change clipboard&quot;,&quot;lynda.premium_video&quot;:&quot;PREMIUM VIDEO&quot;,&quot;share&quot;:&quot;Share&quot;,&quot;clipping.toast.share_clip&quot;:&quot;Share clip&quot;,&quot;clipping.select_clipboard_modal.select_another_clipboard&quot;:&quot;Select another clipboard&quot;,&quot;select&quot;:&quot;Select&quot;,&quot;clipping.select_clipboard_modal.select_new_clipboard&quot;:&quot;Select a new clipboard&quot;,&quot;lynda.tld&quot;:&quot;.com&quot;,&quot;right_rail.recommended&quot;:&quot;Recommended&quot;,&quot;views.other&quot;:&quot;views&quot;}" name="ss-i18n-translations" />

  <script type="text/javascript">
    $.extend(slideshare_object, {"key":false,"totalSlides":61,"preview":"no","presentationId":58792278,"jsplayer":{"use_ssl":false,"show_related_content":"1","toolbar_html":"\n\u003C!-- using div.bar-[top, bottom]-margin to fix toolbar spacing with a taller progressbar (improve slide scrubbing UX) --\u003E\n\u003Cdiv class=\"j-progress-bar progress-bar-wrapper\"\u003E\n  \u003Cdiv class=\"progress-bar-spacing\"\u003E\u003C/div\u003E\n  \u003Cdiv class=\"buffered-bar\"\u003E\u003C/div\u003E\n  \u003Cdiv class=\"j-slides-loaded-bar progress-bar\"\u003E\u003C/div\u003E\n  \u003Cdiv class=\"j-progress-tooltip progress-tooltip\" style=\"display: none;\"\u003E\n    \u003Cdiv class=\"j-tooltip-content progress-tooltip-wrapper\"\u003E\n      \u003Cimg class=\"j-tooltip-thumb tooltip-thumb\" onerror=\"this.src=''\"\n      \u003E\n      \u003Cspan class=\"j-slidecount-label slidecount-label\"\u003E1\u003C/span\u003E\n    \u003C/div\u003E\n    \u003Cdiv class=\"progress-tooltip-caret\"\u003E\u003C/div\u003E\n  \u003C/div\u003E\n\u003C/div\u003E\n\u003Cdiv class=\"progress-bar-spacing\"\u003E\u003C/div\u003E\n\n\n\n\u003Cdiv class=\"j-tools bot-actions\"\u003E\n\u003C/div\u003E\u003C!-- .bot-actions --\u003E\n\n\n  \u003Cdiv class=\"j-tools bot-actions\"\u003E\n    \u003Ca data-tooltip\n      aria-haspopup=\"true\"\n      style=\"display: none\"\n      class=\"j-tooltip j-download action-download has-tip\"\n      title=\"Save this \"\n      href=\"/login?from_source=%2FToruUenoyama%2Ftensorflow-gdg%3Ffrom_action%3Dsave\u0026amp;from=download\u0026amp;layout=foundation\"\n      data-target=\"#login_modal\"\n      data-placement=\"top\"\u003E\n      \u003Ci class=\"fa fa-download fa-lg\" style=\"margin-top: 1px;\"\u003E\u003C/i\u003E\n    \u003C/a\u003E\n  \u003C/div\u003E\n\n\n\n\n  \u003Cdiv class=\"nav\"\u003E\n      \u003Cbutton id=\"btnPrevious\" title=\"Previous Slide\"\u003E\n        \u003Cdiv class=\"j-prev-btn arrow-left disabled\"\u003E\u003C/div\u003E\n      \u003C/button\u003E\n    \u003Clabel class=\"goToSlideLabel\"\u003E\n      \u003Cspan id=\"current-slide\" class=\"j-current-slide\"\u003E1\u003C/span\u003E of \u003Cspan id=\"total-slides\" class=\"j-total-slides\"\u003E61\u003C/span\u003E\n    \u003C/label\u003E\n      \u003Cbutton id=\"btnNext\" title=\"Next Slide\"\u003E\n        \u003Cdiv class=\"j-next-btn arrow-right\"\u003E\u003C/div\u003E\n      \u003C/button\u003E\n  \u003C/div\u003E\n\n\n\n\u003Cdiv class=\"navActions\"\u003E\n\n\n\n    \u003Cbutton id=\"btnFullScreen\" class=\"j-tooltip btnFullScreen\" title=\"View Fullscreen\"\u003E\n      \u003Cspan class=\"fa fa-stack\"\u003E\n        \u003Ci class=\"fa fa-expand fa-stack-1x\"\u003E\u003C/i\u003E\n      \u003C/span\u003E\n    \u003C/button\u003E\n    \u003Cbutton id=\"btnLeaveFullScreen\" class=\"j-tooltip btnLeaveFullScreen\" title=\"Exit Fullscreen\"\u003E\n      \u003Cspan class=\"fa-stack\"\u003E\n        \u003Ci class=\"fa fa-compress fa-stack-1x\"\u003E\u003C/i\u003E\n      \u003C/span\u003E\n    \u003C/button\u003E\n\n\u003C/div\u003E\n\n","fullscreen_bgcolor":"jsplBgColorBlack","bambooleaf_presentation":false,"slideview_url":"/ToruUenoyama/tensorflow-gdg","disable_eagerload":true,"related_position":0,"preload_after_pageload":true,"mode":"image","slide_error_template":"\u003Cstyle type=\"text/css\"\u003E\n  .jsplayer-slide-error {\n    background-color: #000;\n    padding: 20% 0 13% !important;\n  }\n  .jsplayer-slide-error div {\n    text-align: center;\n  }\n  .jsplayer-slide-error img {\n    height: 79px !important;\n    margin: 0 0 10px;\n    width: 80px !important;\n  }\n  .jsplayer-slide-error .slide-error-body {\n    color: #eee;\n    font-family: 'Lucida Grande',Verdana, Arial, Helvetica, sans-serif;\n    padding: 0 !important;\n  }\n  .jsplayer-slide-error .slide-error-body p {\n    font-size: 0.8em;\n    line-height: 1.1em;\n    margin: 7px 0 11px;\n  }\n  .jsplayer-slide-error .slide-error-body input[type=button] {\n    margin: 7px 0 0;\n    padding: 7px 14px;\n  }\n\u003C/style\u003E\n\u003Cdiv class=\"jsplayer-slide-error\"\u003E\n  \u003Cdiv style=\"position:relative;\"\u003E\n    \u003Cimg src='//public.slidesharecdn.com/b/images/ssplayer/error_dudes-80x79.png' height=\"79\" width=\"80\" /\u003E\n    \u003Cdiv class=\"slide-error-body\"\u003E\n      \u003Cp\u003EWe were unable to load the slide.\u003C/p\u003E\n      \u003Cinput class=\"btn btn-large\" type=\"button\" value=\"Reload slide\" /\u003E\n    \u003C/div\u003E\n  \u003C/div\u003E\n\u003C/div\u003E\n","bucket_location":"//html.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/","author_id":32207858,"video_slides_count":0,"player_bgcolor":"jsplBgColorBigfoot","image_bucket_location":"//image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739","pin_image_url":"http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-1-638.jpg?cb=1456578209","slide_titles":[],"timestamp":1456578209,"player_type":"presentation","start_slide":1,"hosted_in":"slideview","id":58792278,"is_only_private":false,"spinner_url_fullscreen":"//public.slidesharecdn.com/b/images/ssplayer/loading_black.gif?468e48bc3a","replayscreen":{"html":"\u003Ca href=\"#\" class=\"j-replay-button replay-button\" \u003E\n  \u003Ci class=\"fa fa-refresh\"\u003E\u003C/i\u003E\n  View again\n\u003C/a\u003E\n"},"bambooleaf_hash":false,"iframe_url":"//www.slideshare.net/slideshow/embed_code/key/tkPJMfS5uuYI0N","share_text":"TensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a) by Toru UENOYAMA via slideshare","disable_lastscreen":false,"twitter_recommended_users":null,"has_video":false,"lastscreen":{"related":[{"title":"Introduction to Information Archite...","author_login":"mikecrabb","views":5710,"url":"/mikecrabb/introduction-to-information-architecture-58750805","author":"Mike Crabb","thumbnail":"//cdn.slidesharecdn.com/ss_thumbnails/introductiontoinformationarchitecture-160226103935-thumbnail.jpg?cb=1456521962"},{"title":"The Link Between Alcohol and Breast...","author_login":"omerhameed","views":2577,"url":"/omerhameed/the-link-between-alcohol-and-breast-cancer","author":"Dr.  Omer Hameed","thumbnail":"//cdn.slidesharecdn.com/ss_thumbnails/risk-of-breast-cancer-160226020800-thumbnail.jpg?cb=1456521687"},{"title":"The Death of Polling?","author_login":"IpsosMORI","views":2665,"url":"/IpsosMORI/the-death-of-polling","author":"Ipsos MORI","thumbnail":"//cdn.slidesharecdn.com/ss_thumbnails/thedeathofpolling-160226121628-thumbnail.jpg?cb=1456521635"},{"title":"Publishing Production: From the Des...","author_login":"DeantaGlobal","views":1777,"url":"/DeantaGlobal/publishing-production-from-the-desktop-to-the-cloud","author":"Deanta ","thumbnail":"//cdn.slidesharecdn.com/ss_thumbnails/desktoptothecloud-160226132022-thumbnail.jpg?cb=1456521576"},{"title":"Mobile App Prototyping for Designers","author_login":"jonoYoung","views":2296,"url":"/jonoYoung/1280x800geekendprotypingfordesigners","author":"Jono Young","thumbnail":"//cdn.slidesharecdn.com/ss_thumbnails/0c780e37-2098-4e84-971b-160c56b9f3ec-160226200020-thumbnail.jpg?cb=1456521500"},{"title":"How Email and Great Content Can Fue...","author_login":"visually","views":2447,"url":"/visually/how-email-and-great-content-can-fuel-your-buyers-journey","author":"Visually","thumbnail":"//cdn.slidesharecdn.com/ss_thumbnails/visuallyemmadeck-edit-3-160226193745-thumbnail.jpg?cb=1456521484"}],"url":"http://www.slideshare.net/ToruUenoyama/tensorflow-gdg"},"fullscreen_url":"/fullscreen/ToruUenoyama/tensorflow-gdg","sharescreen":{"title":"TensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GD...","user_name":"","url":"http://www.slideshare.net/ToruUenoyama/tensorflow-gdg","slideshow_id":58792278,"html":"\u003Cdiv class=\"shareScreen\"\u003E\n  \u003Ca href=\"#\" class=\"close\"\u003E\u0026times;\u003C/a\u003E\n  \u003Cul class=\"shareMethods\"\u003E\n    \u003Cli class=\"embed\"\u003E\n      \u003Clabel for=\"embed-code\"\u003EEmbed\u003C/label\u003E\n      \u003Cinput type='text' class=\"shareScreenEmbedCode\" value=\"code\" name=\"embed-code\" /\u003E\n    \u003C/li\u003E\n    \u003Cli class=\"url last\"\u003E\n      \u003Clabel for=\"embed-url\"\u003EURL\u003C/label\u003E\n      \u003Cinput type='text' class=\"shareScreenSSUrl\" value=\"code\" name=\"embed-url\" /\u003E\n    \u003C/li\u003E\n  \u003C/ul\u003E\n  \u003Cform class=\"emailShare\"\u003E\n    \u003Cfieldset\u003E\n      \u003Clegend\u003EEmail this\u003C/legend\u003E\n      \u003Cinput type=\"hidden\" class=\"shareDefaultMessage\" value=\"I think you will find this useful.\" /\u003E\n      \u003Cul\u003E\n        \u003Cli\u003E\n          \u003Clabel for=\"name\"\u003EYour name\u003C/label\u003E\n          \u003Cinput class='shareScreenFromName' type=\"text\" value=\"\" /\u003E\n        \u003C/li\u003E\n        \u003Cli\u003E\n          \u003Clabel for=\"mailID\"\u003EEmail to\u003C/label\u003E\n          \u003Cinput class=\"shareScreenMailID\" type=\"text\" value=\"\" /\u003E\n        \u003C/li\u003E\n        \u003Cli class=\"submit\"\u003E\n          \u003Clabel\u003E\u0026nbsp;\u003C/label\u003E\n          \u003Cinput class=\"shareSprite\" type=\"submit\" value=\"\" /\u003E\n        \u003C/li\u003E\n      \u003C/ul\u003E\n      \u003C/fieldset\u003E\n  \u003C/form\u003E\n\u003C/div\u003E\u003C!-- shareScreen ends here --\u003E\n"},"container":"svPlayerId","render_links":"default","next_slideshow_pos":null,"embed_sizes":{"config":{"defaultPreset":"preset4"},"presets":{"preset2":{"size":{"height":355,"width":425},"displaySize":{"height":38,"width":50}},"preset3":{"size":{"height":420,"width":510},"displaySize":{"height":45,"width":60}},"preset4":{"size":{"height":485,"width":595},"displaySize":{"height":53,"width":70}},"preset1":{"size":{"height":290,"width":340},"displaySize":{"height":30,"width":40}}}},"image_ready":true,"bambooleaf_enabled":null,"show_image_player":true,"spinner_url":"//public.slidesharecdn.com/b/images/ssplayer/loading_bigfoot.gif?8d8fb5905f","iframe_code":"\u003Ciframe src=\"{iframe_url}\" width=\"{width}\" height=\"{height}\" frameborder=\"0\" marginwidth=\"0\" marginheight=\"0\" scrolling=\"no\" style=\"border:1px solid #CCC; border-width:1px; margin-bottom:5px; max-width: 100%;\" allowfullscreen\u003E \u003C/iframe\u003E \u003Cdiv style=\"margin-bottom:5px\"\u003E \u003Cstrong\u003E \u003Ca href=\"https://www.slideshare.net/ToruUenoyama/tensorflow-gdg\" title=\"TensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)\" target=\"_blank\"\u003ETensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)\u003C/a\u003E \u003C/strong\u003E from \u003Cstrong\u003E\u003Ca target=\"_blank\" href=\"http://www.slideshare.net/ToruUenoyama\"\u003EToru UENOYAMA\u003C/a\u003E\u003C/strong\u003E \u003C/div\u003E","page":1,"disable_share":false,"twitter_share_text":"TensorFlow \u3092\u4f7f\u3063\u305f \u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a) #machinelearning","ppt_location":"machinelearningwithtensorflow20160227-160227125739","stripped_title":"tensorflow-gdg","html_ttffont_url_suffix":".js","next_prev_experiment":true,"slide_count":61,"rel_slide_urls":[],"meta_error_template":"\u003Cstyle type=\"text/css\"\u003E\n  .jsplayer-slide-error {\n    background-color: #000;\n    padding: 20% 0 13% !important;\n  }\n  .jsplayer-slide-error div {\n    text-align: center;\n  }\n  .jsplayer-slide-error img {\n    height: 79px !important;\n    margin: 0 0 10px;\n    width: 80px !important;\n  }\n  .jsplayer-slide-error .slide-error-body {\n    color: #eee;\n    font-family: 'Lucida Grande',Verdana, Arial, Helvetica, sans-serif;\n    padding: 0 !important;\n  }\n  .jsplayer-slide-error .slide-error-body p {\n    font-size: 0.8em;\n    line-height: 1.1em;\n    margin: 8px 0;\n  }\n  .jsplayer-slide-error .slide-error-body input[type=button] {\n    margin: 7px 0 0;\n    padding: 7px 14px;\n  }\n\u003C/style\u003E\n\u003Cdiv class=\"jsplayer-slide-error\"\u003E\n  \u003Cdiv style=\"position:relative;\"\u003E\n    \u003Cimg src='//public.slidesharecdn.com/b/images/ssplayer/error_dudes-80x79.png' height=\"79\" width=\"80\" /\u003E\n    \u003Cdiv class=\"slide-error-body\"\u003E\n      \u003Cp\u003EWe have encountered an error.\u003C/p\u003E\n      \u003Cp\u003EPlease refresh the page.\u003C/p\u003E\n    \u003C/div\u003E\n  \u003C/div\u003E\n\u003C/div\u003E\n","is_private":false,"inpage_full_screen":true,"autoplayOnEmbed":false,"embed_code":"\u003Ciframe src=\"https://www.slideshare.net/slideshow/embed_code/key/tkPJMfS5uuYI0N\" width=\"597\" height=\"486\" frameborder=\"0\" marginwidth=\"0\" marginheight=\"0\" scrolling=\"no\" style=\"border:1px solid #CCC; border-width:1px; margin-bottom:5px; max-width: 100%;\" allowfullscreen\u003E \u003C/iframe\u003E \u003Cdiv style=\"margin-bottom:5px\"\u003E \u003Cstrong\u003E \u003Ca href=\"https://www.slideshare.net/ToruUenoyama/tensorflow-gdg\" title=\"TensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)\" target=\"_blank\"\u003ETensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)\u003C/a\u003E \u003C/strong\u003E from \u003Cstrong\u003E\u003Ca target=\"_blank\" href=\"http://www.slideshare.net/ToruUenoyama\"\u003EToru UENOYAMA\u003C/a\u003E\u003C/strong\u003E \u003C/div\u003E","track_slide_enable":1,"html_eotfont_url_suffix":"-eot.js","wp_code":"[slideshare id=58792278\u0026doc=machinelearningwithtensorflow20160227-160227125739]"},"doc":"machinelearningwithtensorflow20160227-160227125739","downloads":{"allow":true,"sp_isdwnl":true},"fb_app_name":"slideshare","comments":{"captcha_url":"http://s3.amazonaws.com/ss-captchas/","ajaxurl":"/~/slideshow/comments/58792278.json","total_count":0},"asset_id":"7c4cd8d525165ef63899ad9317a2180252ea58a4","show_branding":1,"activities":{"favorites":{"total":6,"url":"/~/slideshow/favorites_list/58792278.json","count":5}},"stats":{"url":"http://www.slideshare.net/~/slideshow/stats/58792278.json"},"slideshow_placeholder":"//public.slidesharecdn.com/b/images/thumbnail.png","top_nav":{"get_url":"/top_nav"},"li_bar":{"get_url":"/li_bar"},"facebook_app_id":"2490221586","embeds_count":0,"startSlide":1,"default_tab":".svMoreAuthor","bizo_partner_id":870,"version_no":"1456578209","user":{"clipboards":null,"clipboards_url":null,"member_type":"non-member","clips_number":0},"li_sponsored_content_related":{"placeholders":[{"id":"sponsored-content-1","position":0}],"url":"https://www.linkedin.com/csp/javascript"},"flagging":{"flagged_value":null},"pvt":0,"analytics_api_enabled":true,"userimage_placeholder":"//public.slidesharecdn.com/b/images/user-48x48.png","relative_static_origin_server":"//public.slidesharecdn.com/b/","slideshow":{"user_login":"ToruUenoyama","title":"TensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)","ss_url":"http://www.slideshare.net/ToruUenoyama/tensorflow-gdg","user_name":"Toru UENOYAMA","pin_image_url":"http://cdn.slidesharecdn.com/ss_thumbnails/machinelearningwithtensorflow20160227-160227125739-thumbnail-4.jpg?cb=1456578209","is_clippable":true,"id":"58792278","type":"presentation","iframe_url":"https://www.slideshare.net/slideshow/embed_code/key/tkPJMfS5uuYI0N","fullscreen_bg_color":"Black","zeroclipboard_url":"http://static.slidesharecdn.com/ZeroClipboardv2.swf","embed_sizes":{"config":{"defaultPreset":"preset4"},"presets":{"preset2":{"size":{"height":355,"width":425},"displaySize":{"height":38,"width":50}},"preset3":{"size":{"height":420,"width":510},"displaySize":{"height":45,"width":60}},"preset4":{"size":{"height":485,"width":595},"displaySize":{"height":53,"width":70}},"preset1":{"size":{"height":290,"width":340},"displaySize":{"height":30,"width":40}}}},"allow_embeds":true,"lead_form_url":"https://www.slideshare.net/slideshow/tkPJMfS5uuYI0N/lead-form","is_clickgen_enabled":false,"iframe_code":"\u003Ciframe src=\"{iframe_url}\" width=\"{width}\" height=\"{height}\" frameborder=\"0\" marginwidth=\"0\" marginheight=\"0\" scrolling=\"no\" style=\"border:1px solid #CCC; border-width:1px; margin-bottom:5px; max-width: 100%;\" allowfullscreen\u003E \u003C/iframe\u003E \u003Cdiv style=\"margin-bottom:5px\"\u003E \u003Cstrong\u003E \u003Ca href=\"https://www.slideshare.net/ToruUenoyama/tensorflow-gdg\" title=\"TensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)\" target=\"_blank\"\u003ETensorFlow \u3092\u4f7f\u3063\u305f\u000B\u6a5f\u68b0\u5b66\u7fd2\u3053\u3068\u306f\u3058\u3081 (GDG\u4eac\u90fd \u6a5f\u68b0\u5b66\u7fd2\u52c9\u5f37\u4f1a)\u003C/a\u003E \u003C/strong\u003E from \u003Cstrong\u003E\u003Ca target=\"_blank\" href=\"http://www.slideshare.net/ToruUenoyama\"\u003EToru UENOYAMA\u003C/a\u003E\u003C/strong\u003E \u003C/div\u003E","view_action_state":"unpublished","allow_clips":true,"social_urls":{"google":"https://plus.google.com/share?url=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg","twitter":"https://twitter.com/intent/tweet?via=SlideShare\u0026text=TensorFlow+%E3%82%92%E4%BD%BF%E3%81%A3%E3%81%9F+%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%81%93%E3%81%A8%E3%81%AF%E3%81%98%E3%82%81+%28GDG%E4%BA%AC%E9%83%BD+%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E5%8B%89%E5%BC%B7%E4%BC%9A%29+%23machinelearning+http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg","facebook":"https://facebook.com/sharer.php?u=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg\u0026t=TensorFlow+%E3%82%92%E4%BD%BF%E3%81%A3%E3%81%9F%0B%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%81%93%E3%81%A8%E3%81%AF%E3%81%98%E3%82%81+%28GDG%E4%BA%AC%E9%83%BD+%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E5%8B%89%E5%BC%B7%E4%BC%9A%29","linkedin":"https://www.linkedin.com/cws/share?url=http%3A%2F%2Fwww.slideshare.net%2FToruUenoyama%2Ftensorflow-gdg\u0026trk=SLIDESHARE"},"recommendations":{"finalRankerModel":"model_001","designKey":"design_001"},"total_slides":61,"clickgen_url":null,"mobile_app_url":"slideshare-app://ss/58792278","is_private":false,"facade_slide_url":"http://image.slidesharecdn.com/machinelearningwithtensorflow20160227-160227125739/95/tensorflow-gdg-1-638.jpg?cb=1456578209","is_author_premium":false,"clips":{},"clip_counts":{},"wp_code":"[slideshare id=58792278\u0026doc=machinelearningwithtensorflow20160227-160227125739]"},"useHttp":1,"dev":false,"stripped_title":"tensorflow-gdg","category":{"featured":0},"gam_cat_name":"technology","is_free_author":false});
  </script>

    <script src="http://public.slidesharecdn.com/b/ss_foundation/combined_player_presentation_init.js?86d81f83fb" type="text/javascript"></script>

    <script type="text/javascript">
      $(document).ready(function() {
        var $el = $('#svPlayerId');
        var classMap = {
          'document': 'document_player',
          'html': 'html_player',
          'infographic': 'infographic_player',
          'video': 'video_player'
        }

        player = new SSPlayer(slideshare_object.jsplayer);
        $(player).bind('slidechanged', function(e) {
          if (typeof(loadDataForSlide) === 'function') {
            loadDataForSlide(e.ssData.index);
          }
        });

        $el.addClass(classMap[player.config.player_type]);
      });
    </script>



  <script>

      slideshare_object.add_signin_link('.j-favorite');
      slideshare_object.add_login_source('.j-favorite', 'slideview.top_toolbar.like');
      slideshare_object.add_login_source('.j-save', 'slideview.top_toolbar.download');
      slideshare_object.addSigninFrom('.j-favorite', 'favorite');
      slideshare_object.bindToModalLogin('.j-favorite');
      slideshare_object.bindToModalLogin('.j-save');

    slideshare_object.bind_favorites('#slideview-container');

    $(document).ready(function() {

      // Favorites update

      // Remove some sidebar list items to limit the sidebar's height
      var mainContentHeight = $("#main-panel").height();
      var sidePanelItemHeight = $("#side-panel .j-related-item").first().outerHeight();
      var numItemToDisplay = Math.floor((mainContentHeight - 200) / sidePanelItemHeight);

      $("#side-panel .tabs-content .content").each(function() {
        $(this).find(".j-related-item").slice(numItemToDisplay - 2).remove();
      });

        var loadAdditionalFunctionality = function() {
          e = document.createElement('script');
          e.type = 'text/javascript';
          e.async = true;
          e.src =  '//public.slidesharecdn.com/b/ss_foundation/combined_slideview_loggedout.js?7c4cd8d525165ef63899ad9317a2180252ea58a4';
          var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(e, s);
        };

        $(window).load(function() {
          loadAdditionalFunctionality();
        });


      // Load additionnal css
      $(window).load(function() {
        loadCSS("//public.slidesharecdn.com/b/ss_foundation/stylesheets/slideview.css?7c4cd8d525165ef63899ad9317a2180252ea58a4");
      });
    });
  </script>

  

  <noscript>
    <img height="1" width="1" alt="" style="display:none;"
      src="//www.bizographics.com/collect/?pid=870&fmt=gif" alt="Bizographics tracking image" />
  </noscript>


  </body>
</html>
