In a broad sense, can all the deep learning tasks be viewed as classification?
1 Answer
No.
It is totally reasonable to predict values that exist on a continuum (“regression”). In such a scenario, it makes no sense to assign a probability to any particular value, since every value taken by a continuous distribution has zero probability.
Further, a classical classification problem gives no sense of order among the categories. If the predictions are real numbers, then not only did a model predicting $4$ when the truth is $7$ get it wrong, but the model is more wrong than it would have been had it predicted $5$.