Emmanouil-Vasileios Vlatakis Gkaragkounis : Bridging the Gap between Theory and Practice: Solving Intractable Problems in a Multi-Agent Machine Learning World

Posted by Lev Reyzin , part of the Departmental Colloquium.

At

Jan. 22, 2024, 3 p.m.

In

636 SEO

Abstract

Traditional computing sciences have made significant strides using tools like Complexity and Worst-Case Analysis over the past decades. However, the rise of Machine Learning has brought a renewed focus on complex optimization problems found in diverse areas such as RoboSoccer, image generation, autonomous vehicles, and multi-objective logistics optimization. While theoretically challenging, these problems often prove more manageable in real-world scenarios, thanks to modern Machine Learning techniques which surprisingly often rely on straightforward methods like Local Search and Gradient Descent. In this talk, I'll explore why these seemingly simple algorithms are effective in complex environments. We'll look into developing a theory that moves beyond traditional analysis, connecting theoretical concepts with practical applications. The discussion will also cover decision-making in scenarios with conflicting incentives and uncertainty, using advanced methods from Optimization, Statistics, and Game Theory. We'll delve into the dynamics of strategic decision-making where data uncertainty and opposing incentives play a crucial role.