Solvable Models of Scaling and Emergence in Deep Learning

Explore key insights into deep learning through a 57-minute lecture from Harvard University's Cengiz Pehlevan, presented at the CMSA Mathematics and Machine Learning Closing Workshop. Delve into mathematical models that explain scaling behaviors and emergent properties in deep learning systems, examining how these theoretical frameworks help understand the fundamental principles governing neural networks. Learn about solvable models that bridge the gap between theoretical predictions and empirical observations in machine learning, offering valuable perspectives on the mathematical foundations underlying modern AI systems.