JS
On the scientific method and its application to the science of deep learning
Backsolving classical generalization bounds from the modern kernel regression eigenframework
One kernel, many eigensystems
A complete characterization of the expressivity of shallow, bias-free ReLU networks
The eigensystem of the Gaussian kernel w.r.t. a Gaussian measure
The optimal low-rank solution for linear regression
Infinite-width autoencoders are cursed
It's hard to turn a low-rank matrix into a high-rank matrix
Insights into GPT-2's positional encodings
Understanding fractals from iterated maps