Minimizing Power for Neural Network Training with Logarithm-Approximate Floating-Point Multiplier | IEEE Conference Publication | IEEE Xplore