Prof. Ofer Shir

The advent of modern laboratory and field experiments, as well as of computerized systems, enables researchers to control experiments and analyze their big-data in high-speed rates. Yet, algorithmic design of intelligent experiments for optimization and learning of prescribed objectives may constitute the next level. Very little is known today about the general usefulness of optimization heuristics and statistical learning methods in laboratory experiments, about their strengths and weaknesses when compared to traditional Design-of-Experiments approaches, or about any kind of guidelines which approach to prefer, depending upon the dimensionality of the search-space, the levels of uncertainty and the number of available trials.

We are interested in learning and optimization questions related to systems within the Natural Sciences. In particular, Operational Research and Machine Learning aspects of experimental domains, that is, practical physical- and life-sciences problems whose computerized models are unavailable or too expensive to be executed, and thus necessitate real-world experiments toward the end of their global optimization. Our research interests encompass in general the following topics:

• AI-Driven Scientific Research and Algorithmically-Guided Experimentation
• Artificial General Intelligence, Deep Learning, Self-Supervised Learning
• Combinatorial Optimization (White/Grey/Black-box, including physical systems in the lab) and Integer Evolution Strategies
• Quantum Optimization and Quantum Control

Our current activities include Experimental Combinatorial Optimization (with Dan Gamrasni), Mixed-Integer Evolution Strategies (with Thomas Bäck and Michael Emmerich), and Benchmarking Quantum Computers on Combinatorial Optimization Problems (with Alberto Moraglio).