crazyn2
commented
8 months ago Not only that, the memory module does not align with the description in the paper. The ROC score is significantly influenced by the batch size; if the batch size is not 1, the results would deteriorate.
In the description of section 3.1.1, "Differently, we remove the last batch normalization [12] and ReLU layers [18] in the encoder, as the ReLU cuts off negative values, restricting diverse feature representations. We instead add an L2 normalization layer to make the features have a common scale" But, the code just removes the BN and Relu Layres, no add L2 normalization layer? how to do this add operation?