Infinite Limits of Neural Networks and A Dynamical Model of Neural Scaling Laws
Check out these blog posts part one and part two as well as the Tex writeup. These were authored by me, Haozhe Shan, Boaz Barak, Abdul Canatar, Cengiz Pehlevan.
This colab notebook contains an introduction to an average case analysis of generalization in kernel regression, exploring spectral bias and the non-monotonicity of the risk curves in the presence of noise.