Encrypted Collaborative Filtering
One-line Summary
A privacy-preserving recommendation system using encrypted IDs, zero-knowledge proof validation, and matrix factorization.
Problem
Recommendation systems often rely on user-item interaction data, which can expose sensitive user behavior if handled carelessly. This project explores how collaborative filtering can be paired with privacy-preserving mechanisms while still producing useful recommendations.
System Architecture
The system combines:
- MovieLens-style user-item rating data
- Encrypted identifiers for privacy protection
- Matrix factorization for collaborative filtering
- Zero-knowledge proof validation concepts
- Evaluation with recommendation metrics such as RMSE and MAE
My Contribution
- Implemented the recommendation model workflow with PyTorch
- Integrated encryption-oriented handling for identifiers
- Studied how privacy mechanisms interact with recommendation quality
- Evaluated the system with standard collaborative filtering metrics
Technical Stack
Python PyTorch MovieLens AES Zero-Knowledge Proof Matrix Factorization
Results / Current Status
This project is a strong complement to the embedded AI work because it shows machine learning system design through a privacy and security lens.
What I Learned
- How matrix factorization supports collaborative filtering
- Why privacy-preserving system design affects data modeling choices
- How recommendation metrics help evaluate model behavior beyond implementation correctness