Training deep neural networks with low precision multiplications
This addresses power and space efficiency for hardware implementations of neural networks, though it is incremental as it applies existing methods to new precision formats.
The study tackled the problem of reducing computational cost in training deep neural networks by evaluating low-precision multiplications, finding that very low precision, such as 10-bit multiplications, is sufficient for training state-of-the-art networks on benchmark datasets like MNIST, CIFAR-10, and SVHN.
Multipliers are the most space and power-hungry arithmetic operators of the digital implementation of deep neural networks. We train a set of state-of-the-art neural networks (Maxout networks) on three benchmark datasets: MNIST, CIFAR-10 and SVHN. They are trained with three distinct formats: floating point, fixed point and dynamic fixed point. For each of those datasets and for each of those formats, we assess the impact of the precision of the multiplications on the final error after training. We find that very low precision is sufficient not just for running trained networks but also for training them. For example, it is possible to train Maxout networks with 10 bits multiplications.