EE Seminar: On the Information Complexity of Learning
(The talk will be given in English)
Speaker: Ido Nachum
Department of Mathematics, Technion
Monday, April 8th, 2019
15:00 - 16:00
Room 011, Kitot Bldg., Faculty of Engineering
On the Information Complexity of Learning
Abstract
Under an information theoretic setting, these two notions are indeed equivalent.
a) Compression implies learning. We will show that learning algorithms that retain a small amount of information from their input sample generalize.
b) Learning implies compression. We will show that under an average-case analysis, every hypothesis class of finite VC dimension (a characterization of learnable classes) has empirical risk minimizers (ERM) that do not reveal a lot of information.
If time permits, we will discuss a worst-case lower bound by presenting a simple concept class for which every empirical risk minimizer (possibly randomized) must reveal a lot of information and also observe connections with well-studied notions such as sample compression schemes, Occam's razor, PAC-Bayes, and differential privacy.