Peter E. Latham

Convex Efficient Coding

Saddle-to-Saddle Dynamics Explains A Simplicity Bias Across Neural Network Architectures

A Biologically Plausible Dense Associative Memory with Exponential Capacity

Training Dynamics of In-Context Learning in Linear Attention

Range, not Independence, Drives Modularity in Biologically Inspired Representations

When Are Bias-Free ReLU Networks Effectively Linear Networks?