Cryptography for privacy-preserving AI : Challenges and solutions

The goal of Privacy Preserving Machine Learning (PPML) is to identify customized algorithms that would, by design, preserve the privacy of the processed data. Fully homomorphic encryption or secure multi-party computation are popular cryptographic techniques for PPML. Yet, these often incur high computational and/or communication costs. In this talk, we will analyse the tension between ML techniques and relevant cryptographic tools, and overview existing solutions addressing privacy requirements.