Introduction
Most famous deep-learning successes are classification: recognising objects in photos, detecting spam, diagnosing disease from scans, identifying spoken words. An MLP classifier is built just like a regression MLP — dense hidden layers with ReLU — but its output layer and loss are designed to produce class probabilities.
Think of the output layer as a panel of judges, one per class, each holding up a score. The final step turns those scores into probabilities that say how confident the network is in each class.