So I am trying to enforce better separability in my deep learning model and was wondering what I can use besides cross entropy loss to do that? Could maybe using logarithm with different basis in cross entropy (i.e. using lower basis of logarithm than $e$ to gain steeper losses on small values, or bigger basis of logarithm to enforce plateaued losses). What would you suggest on doing?

  • $\begingroup$ What do you mean by "better class separability"? Can you explain maybe with an example? What exactly do you want to achieve? $\endgroup$ – nbro Oct 9 '20 at 14:15
  • $\begingroup$ @nbro I want to force the network to give higher softmax outputs for the class it thinks are the correct ones. i.e. to force it to give instead of for three class example 0.3,0.5,0.2 to give something like 0.2 0.8 0.1, but to avoid doing it in programming, but actually force different loss function that would force this kind of behavior. $\endgroup$ – GreatDuke Oct 11 '20 at 12:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.