Privacy-Preserving Machine Learning Inference with Zero-Knowledge Proofs

Title: Privacy-Preserving Machine Learning Inference with Zero-Knowledge Proofs

Mentor : Dr. Mumin Cebe

In lightweight machine learning frameworks, we can now perform machine learning inference on edge devices without transmitting potentially sensitive data to centralized servers, which improves privacy and scalability. However, protecting the input and model parameters from public view presents a challenge, especially when dealing with sensitive personal data.

To address this challenge, the use of zero-knowledge proofs in combination with machine learning can provide a novel approach that satisfies the seemingly contradictory demands of privacy and correctness. The aim of this REU project is to:

• Understand the principles and protocols of zero-knowledge proofs and how they can be used in machine learning inference. • Investigate the potential of applying zero-knowledge proofs in lightweight machine learning frameworks for privacy-preserving inference on edge devices. • Develop and evaluate algorithms that enable the secure transmission and processing of sensitive data while maintaining privacy and correctness for downstream entities, such as on-chain smart contracts.