Modular Arithmetic Challenges
The discussion delves into the complexities of applying gradient descent in modular arithmetic, particularly when dealing with linear regression on an n-dimensional torus. Kristin highlights the difficulties posed by large step sizes that can lead to overshooting solutions, while also sharing encouraging findings from recent work on circular regression. Despite some promising results, the ongoing debate about the effectiveness of gradient descent in this context suggests that further research is essential.In this clip
From this podcast
The Gradient
Kristin Lauter: Private AI, Homomorphic Encryption, and AI for Cryptography
Related Questions