About event:
We invite you to the sixth meeting of the Student Mathematics Seminar. The topic of the report: “PAC learning and topology”.
The PAC (probably approximately correct learning), proposed by Leslie Valiant back in the 1980s, is one of the fundamental frameworks of machine learning theory. Despite its abstract nature, it remains relevant and still serves as the basis for a large part of ML theoretical research.
At the seminar, we will look at the PAC model from an unexpected but natural angle - from the perspective of topology. In particular, we will discuss the use of topological tools, such as the Borsuk-Ulam theorem, to analyze learning stability. As an example, we will analyze a topologically grounded barrier to list-replicable learning.
The seminar will be especially interesting for those who delve into machine learning theory or work at the intersection of mathematics and ML, regardless of their course of study or university.
Speaker:
Bohdan Chornomaz - PhD, postdoctoral fellow at the Faculty of Mathematics, Technion-Israel Institute of Technology.
Language:
Ukrainian