1b9a80606d74d3da6db2f1274557e644-AuthorFeedback.pdf

Here isanexample (simplified andonlybriefly explained4 for length) in which the loss will be multiscale as considered in our paper: train a2-layer neural network to fit data5 {xk,yk}k, where the outputyk = yk0 +yk1 +ξk admits a decomposition into large scale behavioryk0 = g0(xk),6 microscopic detail...