8-Bit Approximations for Parallelism in Deep Learning

Tim Dettmers

The creation of practical deep learning data-products often requires parallelization across processors and computers to make deep learning feasible on large data sets, but bottlene…