Helen - Maliciously Secure Coopetitive Learning for Linear Models

Explore a groundbreaking system for secure collaborative machine learning in this IEEE conference talk. Dive into Helen, a system enabling multiple organizations to train linear models on combined datasets without compromising privacy or competitive advantage. Learn about the challenges of coopetitive learning, the threat model addressing malicious adversaries, and the innovative techniques employed for secure multiparty computation. Discover how Helen achieves significant performance improvements compared to existing frameworks, and understand its potential applications in fields such as medical research and fraud detection. Gain insights into the system's setup, input preparation, iterative training process, and model release, concluding with an evaluation of its effectiveness in protecting sensitive data while fostering collaborative advancements.

Intro
Fraud detection
Solution?
Concerns
Secure multiparty computation
Desired security properties
Threat model
Prior work
Challenge
Techniques
Setup
Input preparation
Iterative training
Model release
Evaluation
Conclusion