1 paper
Arsalan Sharifnassab, Mohamed Elsayed, Kris De Asis +2
In gradient-based learning, a step size chosen in parameter units does not produce a predictable per-step change in function output. This often leads to instability in the streamin…