Dmitry Vaintrob

Power-law distributions in neural network weight matrices may explain the inductive bias toward sparse representations, according to research presented under the Iliad Fellowship. The generalized cent…