Great post. Long sequences are a problem I have been dealing with long enough now. Even though GRU outperforms LSTM in longer sequences, it still isn't enough a

Great post. Long sequences are a problem I have been dealing with long enough now. Even though GRU outperforms LSTM in longer sequences, it still isn't enough and vanishing gradients problem is still experienced. Currently, one-dimensional convolution units are effective while dealing with longer time steps, but sadly it's not the precise unit people wish to use while processing longer time-steps.