Member-only story
Privacy-Aware Machine Learning: Homomorphic Encryption and Federated Learning
At LastingAsset Limited, we’re working on new models for fraud detection using homomorphic encryption. So let’s have a quick look at some of the current work in the area, especially in how we can integrate with privacy-aware machine learning. Most of the work in this uses the CCKS homomorphic method, and which operates on floating point values. With this, we use a public key to encrypt the data, and where we can then process with this encrypted data in a privacy-preserving way. With the encrypted result, we can then decrypt with the associated private key (Figure 1).
We can easily extend this model, so that we can implement machine learning models, and where Bob, Alice and Peggy provide their data in a privacy-aware way, and where the data processor cannot see the contents of their data:
Homomorphic Encryption and ML
Nugent et al [1] implemented fraud detection with homomorphic encryption using two models…
