A Framework for Acceleration of CNN Training on Deeply-Pipelined FPGA Clusters with Work and Weight Load Balancing | IEEE Conference Publication | IEEE Xplore