DIKU Bits: From Machine Learning to Machine Teaching

Speaker

Bob Pepin, Postdoc in the ML (Machine Learning) section.

Title

From Machine Learning to Machine Teaching

Abstract

Machine Teaching explores the opportunities offered by learning from carefully chosen examples provided by a teacher model, as opposed to examples sampled at random. I will introduce the basic ideas with algorithms for teaching and learning hyperplanes in arbitrary dimension and multivariate Gaussian variables. I will explain how Machine Teaching relates to popular learning methods in LLMs such as in-context learning and test-time compute and finish with open theoretical questions.