Apr 23, 2026
Deep linear networks are a surprisingly useful toy model of weight-space dynamics
Deep linear networks are simple enough to study analytically but rich enough to exhibit key phenomena of neural network training.
Apr 23, 2026
On neural scaling and the quanta hypothesis
Eric J. Michaud
What is the origin of neural scaling laws? What do they tell us about the structure of data? What are the limits of interpretability?
Apr 23, 2026
Perspectives on the science of deep learning
Three essays on building theory that matters.
Coming soon...