Weak to Strong Generalization in Random Feature Models

Explore weak to strong generalization phenomena in random feature models through this 54-minute conference talk delivered at the 2025 Mathematical and Scientific Foundations of Deep Learning Annual Meeting hosted by the Simons Foundation. Delve into the theoretical foundations of how machine learning models can generalize from weaker supervision to stronger performance, with specific focus on random feature model architectures. Examine the mathematical principles underlying this generalization capability and understand its implications for deep learning theory. Gain insights into cutting-edge research that bridges the gap between theoretical understanding and practical applications in modern machine learning systems.