recent discussion of running tensorflow/grad. descent on DSP chips: https://news.ycombinator.com/item?id=13379847 https://news.ycombinator.com/item?id=13379847

recent discussion of running tensorflow/grad. descent on DSP chips: https://news.ycombinator.com/item?id=13379847 https://news.ycombinator.com/item?id=13379847 This is a good writeup on mobile/low power devices, lower precision arithmetic, compression, sparsity, hogwild type "naive" assumptions etc https://arxiv.org/abs/1612.07625 https://arxiv.org/abs/1612.07625