# Supervised Learning It is the most popular and well established learning paradigm. The goal is to learn a good approximation of an unknown function $f$ that maps an input $x$ to an output $y$ with: - Loss function: a function which says "how much is good the approximation (called $h$) of $f$?" - Hypothesis space: a subset of the set of all possible function $f$ Supervised learning can be used every time we can't clearly explain which is this function. Why shouldn't we know what this function looks like? - human cannot perform the task (DNA analysis) - human cannot explain a clear algorithm (medical image analysis) - the task continues to change over time (stocks price prediction) - the task is user-specific (recommender system) Supervised Learning is divided into: - [Linear Regression](01.Linear%20Regression.md) - [Linear Classification](02.Linear%20Classification.md)