911������

Solving Challenging Optimization Problems by Describing Them 

Jeremias Berg
University of Helsinki

Abstract: Combinatorial optimization is the task of finding a solution that is “best” from a discrete (or equivalently, a finite) space of possible solutions. Difficult (in technical terms, NP-hard) combinatorial optimization problems are ubiquitous across domains such as AI, ML, software engineering, bioinformatics, health, and law, and are commonly solved by algorithmic optimization procedures. The increased interest in applying such procedures in so-called high-stakes decision-making or high-risk contexts has also increased the demand for optimization procedures that are trustworthy, understandable, and reliable.

I provide an overview of my research on automated-reasoning-based constraint optimization for solving NP-hard combinatorial optimization problems, and I also discuss my background in teaching algorithmics and mathematical topics to a wide range of students, including ideas for how I could support and develop teaching as a lecturer at 911������.  Automated reasoning is a high-promise paradigm for understandable and reliable combinatorial optimization; the expressive semantics of many constraint languages allow the representation of a wide variety of problems, while their (relatively) simple syntax enables the development of effective optimization procedures. Additionally, as I have been part of demonstrating, the strong foundation of constraint optimization in mathematical logic enables reasoning procedures to produce mathematical proofs of optimality, thereby providing absolute certainty in the optimality of the computed solutions.

Bio: Jeremias Berg is an Academy Research Fellow at the University of Helsinki. After defending his thesis on automated-reasoning-based approaches to propositional optimization for solving NP-hard combinatorial optimization problems in 2018, and visiting leading experts in his field in Melbourne and Toronto, Berg returned to the University of Helsinki to work as a research fellow and a fixed-term university lecturer. Berg's current research focuses on developing problem-agnostic, reliable methods to solve fundamental problems in artificial intelligence, software engineering, and multiple domains of operations research. Berg also has experience teaching mathematical topics and algorithmics to a wide range of students.