I see how the probability (softmax output) contains more info than simply the class with highest probability, but not what you want regarding "significant discr

I see how the probability (softmax output) contains more info than simply the class with highest probability, but not what you want regarding "significant discrimination". Perhaps you want to weight different types of errors differently? Eg: https://github.com/Hezi-Resheff/paper-log-bilinear-loss https://github.com/Hezi-Resheff/paper-log-bilinear-loss