Misbehaving losses while training RawNet1 about rawnet HOT 2 CLOSED

Hi, first of all, this tendency is different to what I observed before.
I recall all three losses decreased in the end.

Are the plots show training until the last epoch?
Since you mentioned the total-loss is decreasing, I'm thinking that the center loss might start to decrease quiet after the starting point.

My suggestion would be training the model using more epochs or smaller batch size to increase the number of total iterations.
Looking at the result of using C_lambda=5, I think with proper balance and enough iterations, you can get low values for all losses

from rawnet.

No, these plots are not for the last epoch. They are only for the initial few epochs.

Will try to train it for some more epochs with smaller batch-size to see if both the losses decrease.

from rawnet.