AI4Sci

T. Tušar, B. Filipič, J. Cork, A. Andova

The Artificial Intelligence for Science (Al4Sci) project is an ambitious initiative designed to harness the power of artificial intelligence (Al) to support and automate scientific processes. For nearly half a century, Al has been envisioned as a tool to assist in the discovery of scientific laws, handling both data (experimental or observational) and knowledge structures such as models, laws, and theories. Machine learning, a subfield of Al, has significantly contributed to data-driven science by deriving knowledge from vast amounts of data, revolutionizing the scientific process. Despite these advancements, limitations remain, especially concerning the transparency and interpretability of Al models. This project addresses these challenges by developing Al approaches from machine learning and knowledge representation, tailored specifically for scientific applications across various domains, including physical sciences and engineering and life sciences.

As part of the AI4Sci project, our department focuses on advancing artificial intelligence for constrained multiobjective optimization (CMO) benchmarking. We aim to enable intelligent algorithm selection, configuration, and performance prediction for complex optimization problems with multiple conflicting objectives and constraints. Our work involves developing a diverse and well-characterized suite of CMO benchmark problems, collecting detailed algorithm performance data, designing a comprehensive ontology and knowledge base for CMO benchmarking, and applying AI methods to support data-driven decision-making in optimization.