1 citations · 1 across the 1 of their papers we have counts for
1 paper · 1 filter
John St John
Here I present a small update to the bias-correction term in the Adam optimizer that has the advantage of making smaller gradient updates in the first several steps of training. Wi…