​Nvidia aims to run neural nets faster, more efficiently

As data gets bigger and models grow larger, deep learning is once again "completely gated by hardware." At the VLSI Symposia, Nvidia suggested some ways to address this problem.