A game theoretic approach to Numerical Approximation, Algorithm Design and Learning
This talk will cover interplays between Game Theory, Numerical Approximation, Gaussian Process Regression and Learning. We will illustrate the interface between statistical inference and numerical analysis through problems related to numerical homogenization, operator adapted wavelets, fast solvers, computation with dense kernel matrices. We will also show how this perspective can be applied in machine learning to the design of bottomless networks (Kernel Flows) that are amenable to some degree of analysis (these networks simulate a data driven stochastic flow in the input space and learn kernels capable of generalization from one interpolation point per class).
This talk will cover joint work with F. Schäfer, C. Scovel, T. Sullivan, L. Zhang and G. R. Yoo.