Encrypted Inference Challenges
Kristin discusses the early attempts to approximate activation functions using polynomials and the implications for scaling in deep learning models. She highlights the ongoing exploration of using nonlinear functions and the advancements made by Zheng in running Lama two on homomorphically encrypted data. The conversation emphasizes the importance of addressing practical concerns related to privacy in data, particularly in sensitive fields like healthcare and finance.In this clip
From this podcast
The Gradient
Kristin Lauter: Private AI, Homomorphic Encryption, and AI for Cryptography
Related Questions