Numerical understanding of neural networks: from representation to learning dynamics
Hongkai Zhao
Duke University
Mathematics Department
In this talk I will present both numerical analysis and experiments to understand a few basic computational issues in using neural network to approximate functions: (1) the numerical error that can be achieved given a finite machine precision, (2) the learning dynamics and computation cost to achieve certain accuracy, and (3) structured and balanced approximation. These issues are studied for both approximation and optimization in asymptotic and non-asymptotic regimes.
