Learning Dynamics in the Feature-Learning Regime - Implicit Bias, Robustness, and Low-Rank Adaptation

Explore the mathematical foundations of deep learning through this 57-minute conference talk examining learning dynamics in the feature-learning regime, with focus on implicit bias, robustness, and low-rank adaptation. Delve into advanced theoretical concepts that underpin how neural networks learn and adapt features during training, presented as part of the 2025 Mathematical and Scientific Foundations of Deep Learning Annual Meeting hosted by the Simons Foundation. Gain insights into the mathematical principles governing feature learning, understand the implicit biases that emerge during neural network training, examine robustness properties of learned representations, and discover how low-rank adaptation techniques influence the learning process in deep neural networks.