6 papers
How are linear representations learned? Exact solutions to the dynamics of abstraction
William W. Yang, Andrew M. Saxe, Peter E. Latham
In artificial and biological neural networks, concepts are often encoded as consistent linear directions in representation space. In deep learning, this idea is known as the linear…
Optimal Learning Rate Scaling Depends on Data in Deep Scalar Linear Networks
Yedi Zhang, Peter E. Latham, Leena Chennuru Vankadara +1
In this short note we consider the gradient descent dynamics of deep scalar linear networks, , which enjoy exact time-course solutions for any integer d…
Saddle-to-Saddle Dynamics Explains A Simplicity Bias Across Neural Network Architectures
Yedi Zhang, Andrew Saxe, Peter E. Latham
Neural networks trained with gradient descent often learn solutions of increasing complexity over time, a phenomenon known as simplicity bias. Despite being widely observed across…
Optimal Learning Rate Schedule for Balancing Effort and Performance
Valentina Njaradi, Rodrigo Carrasco-Davis, Peter E. Latham +1
Learning how to learn efficiently is a fundamental challenge for biological agents and a growing concern for artificial ones. To learn effectively, an agent must regulate its learn…
Training Dynamics of In-Context Learning in Linear Attention
Yedi Zhang, Aaditya K. Singh, Peter E. Latham +1
While attention-based models have demonstrated the remarkable ability of in-context learning (ICL), the theoretical understanding of how these models acquired this ability through…
When Are Bias-Free ReLU Networks Effectively Linear Networks?
Yedi Zhang, Andrew Saxe, Peter E. Latham
We investigate the implications of removing bias in ReLU networks regarding their expressivity and learning dynamics. We first show that two-layer bias-free ReLU networks have limi…