Jago Silberbauer

OS Mathematical Physics November 18 14:30

An Introduction to Statistical Learning

This talk gives an introduction to the concept of 'Probably Approximately Correct'-learning or PAC-learning by means of Binary Classification tasks. We use Hoeffding's Inequality to prove PAC-learnability for finite hypothesis spaces. We then take a look at the infinite case and the tools used there to tackle the question of PAC-learnability.