High performance training of deep neural networks using pipelined hardware acceleration and distributed memory | IEEE Conference Publication | IEEE Xplore