FPGA-based training of convolutional neural networks with a reduced precision floating-point library | IEEE Conference Publication | IEEE Xplore