发帖
雷达卡
Accumulated Decoupled Learning with Gradient Staleness Mitigation for
Convolutional Neural Networks
Huiping Zhuang 1 Zhenyu Weng 1 Fulin Luo 1 Kar-Ann Toh 2 Haizhou Li 3 Zhiping Lin 1
Abstract efficiency, the decoupled learning (Jaderberg et al., 2016)
emerges by addressing one or more of these lockings.
Gradient staleness is a major side effect in decou-
pled learning when training convolutional neural ...
京公网安备 11010802022788号
