In multi-class classification, the primary objective of using multi-class loss functions is to evaluate the model's performance by penalizing incorrect predictions through various mechanisms, such as ______ loss, which is particularly effective in optimizing probabilistic outputs.

The chosen loss uses the logarithm of predicted probabilities, so it strongly penalizes wrong guesses and rewards correct ones. Other options are incorrect because Some think squaring the error helps, but this loss treats all errors the same regardless of probability; The idea that taking the absolute difference works is a misconception.