Something about "LossCosine" function

[Sunliangtai]

It seems that there is something wrong about your function "LossCosine", located at /Layers/TAL_pytorch.py. When I train with cosine regularization, the loss is less than zero, which is totally impossible. And I notice that in function "LossCosine", before returning the mean value of loss, you multiply it with minus one, which is different from that in function "LossDis" function.
And in your paper, the math formulation of your Cosine regularization is also multiplied by minus one, which makes the loss value is less than zero. Is there something wrong with me?

[yingrliu]

Cosine similarity ranges from -1 to 1, and the larger value means that the two vectors have closer directions. So multiplying it by -1 means that we want to maximizing the cosine similarity.