Hidden Progress in Deep Learning: SGD Learns Parities Near the Computational Limit

Open in new window