I am implementing a CNN model for image classification where I am learning about loss functions. There are several types of loss functions to determine error. However, how do we find out which function is suitable for a image multi-classification model?
If you have two classes (i.e. binary classification), you should use a binary crossentropy loss.
If you have more than two you should use a categorical crossentropy loss.
For image classification in keras, with multiple categories,
categorical_crossentropy is a good start.