Mean Field Limit in Neural Network Learning

Explore the theoretical aspects of neural networks in this seminar on mean field limits in neural network learning. Delve into the analysis of neural network behavior during training, focusing on the challenges posed by non-convex optimization and large model sizes. Discover how the mean field limit approach can simplify analysis by considering the dynamics of neural networks with a large number of neurons per layer. Learn about the convergence to global optima in neural networks, shedding light on their optimization capabilities despite non-convexity. Examine the mathematical characterization of data representation in a simple autoencoder. Journey through the progression from two-layer neural networks to multilayer cases, drawing analogies from the physics of interacting particles. Gain insights from the speaker's research, which combines elements of physics, engineering, and mathematics to tackle complex problems in neural network theory.