BatchCrypt - Efficient Homomorphic Encryption for Cross-Silo Federated Learning

Explore a conference talk on BatchCrypt, an efficient homomorphic encryption system for cross-silo federated learning. Dive into the challenges of privacy-preserving machine learning across organizations and discover how BatchCrypt significantly reduces encryption and communication overhead. Learn about novel quantization and encoding schemes, as well as gradient clipping techniques that enable secure aggregation of batched gradients. Understand the implementation of BatchCrypt as a plug-in module for FATE, an industrial cross-silo federated learning framework, and examine its impressive performance improvements in training speed and communication efficiency across geo-distributed datacenters.

Introduction
Target
CrossSilo
Data Center Distributed Training
PrivacyPreserving Techniques
BatchCrypt
Aggregation
Quantization
Quantization Requirements
Gradient Clipping
BatchCrypt Implementation
Batch Script Quantization
Computation Breakdown
Communication Overhead
Results
Comparison
Conclusion
Outro