Deep Neural Network Training without Multiplications

Tsuguo Mogami

Is multiplication really necessary for deep neural networks? Here we propose just adding two IEEE754 floating-point numbers with an integer-add instruction in place of a floating-p…