Regarding fixed precision neural networks, it looks like Intel had some fairly mature work on this at ISCA -- fast inference, but have to train elsewhere in the

Regarding fixed precision neural networks, it looks like Intel had some fairly mature work on this at ISCA -- fast inference, but have to train elsewhere in the machine: https://arxiv.org/abs/1906.09889 https://arxiv.org/abs/1906.09889