Results 23 comments of Anatoly Vostryakov

@TigranGalstyan I haven't done it. It looks like residual connections solve the same task as Recurrent Batch Normalization but easier to implement and train.

@webeng residual connections are just a summation of two tensors. The first tensor is input to the N-th layer, the second tensor is input to the N+1 layer. Nothing to...

Thank you. I close the issue