Tweeted By @andrewgwils

Stochastic Weight Averaging in Low Precision Training (SWALP)! Our new ICML paper (with PyTorch code). SWALP can match the performance of full-precision training, even with all numbers quantized down to 8 bits! https://t.co/eTWQlvUMYS pic.twitter.com/QUjhFeigY7

— Andrew Gordon Wilson (@andrewgwils) April 29, 2019