Introducing GravOptAdaptive — a new version of GravOpt that:https://lnkd.in/dw78avM6 7 day trial, Drop-in replacement for Adam/SGD . There is no need to change the model, data, loss function, or pipeline. → The code continues to work — but it is trained faster (in the paid version) or with a trial limitation (in the free version).

Analyzes the behavior of each parameter separately Gives a larger update only when necessary Keeps the stability of parameters that are already moving well Improves accuracy — without increasing the number of iterations