Minimax rates for learning spectral Barron functions by deep ReLU neural networks

Songqiu Ma, Yunfei Yang

We study how well deep neural networks approximate and learn spectral Barron functions. Recent studies have shown that these function classes can be efficiently approximated by sha…