1 paper · 1 filter
Yiwei Zhou, Ziheng Chen
Stochastic gradient Langevin algorithms often use tamed denominators to stabilize superlinear drifts. This paper shows that when the denominator depends on the current stochastic g…