Secure Multi-Party Computation for ML

Building MPC frameworks for privacy-preserving deep learning at JD.com.

Overview

During my internship at JD.com, I worked on building secure multi-party computation (MPC) frameworks for privacy-preserving deep learning. This project enables multiple parties to jointly train machine learning models without revealing their private data.

Technical Approach

MPC Framework

Developed an MPC framework that supports:

  • Secret Sharing: Splitting data across multiple parties
  • Secure Computation: Performing computations on encrypted data
  • Deep Learning Operations: Supporting neural network training and inference

Privacy Guarantees

The framework provides:

  • Cryptographic security against semi-honest adversaries
  • No party learns any information beyond the computation output
  • Efficient protocols for common ML operations

Applications

  • Federated Learning: Training models across organizations without data sharing
  • Financial Services: Joint risk modeling with customer privacy
  • Healthcare: Collaborative medical research with patient protection

Experience

Research Intern, JD.com (Summer 2022)

  • Designed and implemented MPC protocols for deep learning
  • Optimized computation and communication efficiency
  • Contributed to production-ready privacy infrastructure