Computational learning theory is a theoretical branch of machine learning that studies, with mathematical rigor, what it actually means for a computer to "learn" from data, and under what conditions that learning is guaranteed to work. It asks precise questions like how much training data a given algorithm needs to reliably learn a pattern, and how confident you can be that a model's performance on training data will hold up on new, unseen data.
Rather than building or tuning specific models directly, computational learning theory provides the theoretical foundations that explain why many practical machine learning algorithms work the way they do, and what their fundamental limits are. Concepts developed in this field, like VC dimension and PAC (Probably Approximately Correct) learning, underpin a lot of the reasoning behind why certain models generalize well and others are prone to overfitting.
