Mathematical Introduction, Fully Connected Neural Network, Universality of Neural Networks, Properties of Relu Networks, Counting Neural Linear Regions, Approximation of X 2, Approximation Rates For General Functions, Group Invariant and Equivariant Neural Networks, Derivation of Permutation Equivariant Linear Layers, Universality and Separation For Permutation Invariant Neural Networks, Universality and Separation For General Invariant Neural Networks, Graph Neural Networks, Separation of Graph Neural Networks. Learning Outcomes# at The End of The Course The Students Will Be Able To# 1. Prove Classical Theorems Taught in The Course, Such As The Universality of Neural Networks. 2. Critically Assess Contemporary Articles On Approximation Powers Of Neural Networks. 3. Critically Assess Contemporary Articles On Equivariant Neural Networks. 4. Independently Develop Equivariant Architectures.

Faculty: Computer Science
|Undergraduate Studies |Graduate Studies
Pre-required courses

(104134 - Modern Algebra H and 234125 - Numerical Algorithms)

Course with no extra credit

196014 - Deep Learning and Approximation Theory