One Forward is Enough for Neural Network Training via Likelihood Ratio Method.

Jinyang Jiang, Zeliang Zhang, Chenliang Xu, Zhaofei Yu, Yijie Peng

Browse the full ICLR paper archive.