Go Classes | DA Course | Machine Learning

Which of the following are valid reasons why we minimized the average cross-entropy loss rather than the average squared loss?

1. To prevent our parameters from going to infinity for linearly separable data.
2. There is no closed form solution for the average squared loss.
3. To improve the chance that gradient descent converges to a good set of parameters.
4. The cross entropy loss gives a higher penalty to very wrong probabilities.