A 28nm 8-bit Floating-Point Tensor Core based CNN Training Processor with Dynamic Activation/Weight Sparsification | IEEE Conference Publication | IEEE Xplore