1 paper
Ari Pakman, Lior Kreimer, Yakir Berchenko
Modern deep neural networks often contain far more parameters than needed to fit their training data, yet they achieve impressive generalization. A common explanation for this succ…