sent-conv-torch
sent-conv-torch copied to clipboard
How to deal with padding (0 vector) without maskZero in ConvolutionLayer?
I find there is no maskZero for the temporalConvolutionLayer. In this case, if we meet a window of all embeddings are paddings, which means 0 vectors., I think the convolution layer will always give us the bias.
But wouldn't the output be 0? Or we don't need maskZero in Convolution Layer?