Polylogarithmic width suffices for gradient descent to achieve arbitrarily small test error with shallow ReLU networks

Devs

Polylogarithmic width suffices for gradient descent to achieve arbitrarily small test error with shallow ReLU networks | Read Paper on Bytez