How to make my neural network not predict anything, for example if I am training on cat dog classifier, and in the dataset, there is one image, that is neither cat nor dog, image is noise, then instead of making an incorrect prediction, I want neural network to not make any prediction, like unattempted question.
You could treat this class as a special “unknown” class (and train your model to predict it for these noise images) or alternatively you could treat your use case as a multi-label classification, i.e. each class will get a probability and your model might output zero, one or multiple classes for each sample.