Mill Computing, Inc. › Forums › The Mill › Markets › Stochastic rounding for machine learning › Reply To: Stochastic rounding for machine learning
For training 16-bit perceptrons you shouldn’t need crypto-quality random; 16 bits sampled from the middle of a 32-bit LFSR should be fine. What I don’t understand (haven’t read the cites yet) is why regular FP rounding doesn’t work.