בניין מדעי המחשב

Michal Moshkovitz

University of California

תרצה על

Will lecture on

How to Learn Under Constraints?

The goal of classical machine learning is to learn a high-accuracy model from given examples. The learning process has no constraints besides running in a reasonable time relative to the input size. However, these days, as machine learning is utilized in high-stake applications like healthcare and law, learning has to obey several new constraints. In this talk, I will focus on two types of constraints: explainability and bounded memory. I will present the first explainable algorithm for k-means clustering that has provable guarantees. Then I will focus on another constraint -- learning with bounded memory, where I will present a characterization of high-accuracy learning with bounded memory and its equivalence to learning with statistical queries.

Zoom link:  

https://us02web.zoom.us/j/83383478356