Branchynet: Fast inference via early exiting from deep neural networks
paper关键词:提前结束收益:在保持性能的前提下,加速网络推断速度;削减梯度消失现象。评判标准:分类信息熵,信息熵低于设定阈值,则从该退出点退出。(Ateachexitpoint,BranchyNetusestheentropyofaclassificationresult(e.g.,bysoftmax)asameasureofconfidenceintheprediction.Iftheent