LSTM with peephole connections Paper:Learning Precise Timing with LSTM Recurrent Networks 即 w*x + w*h_(t-1) + w*c_(t-1),t-1为下标的,都是初始状态。 torch.nn.GRU: r是reset,重置门 z是update,更新门 n是new,最新门 参数量:LSTM的0.75倍,计算量也是