We are pleased to share that our paper "Bound-Aware Heuristic Discovery with Large Language Models: A 15-Puzzle Case Study" has been published in Springer's Lecture Notes in Computer Science series, as part of the AIxIA 2026 – Advances in Artificial Intelligence volume. The paper was recently presented at the 24th International Conference of the Italian Association for Artificial Intelligence (AIxIA) in picturesque Perugia, Italy, where the deputy head of our laboratory, Csaba Kiss, represented the authors. It was a great honor to present our work at such a unique venue, a 13th-century Franciscan church.

The research investigates whether large language models can discover efficient, lightweight, and interpretable heuristics for heuristic search. Using the classic 15-puzzle as a controlled testbed, we had language models generate candidate heuristic functions and then evaluated them under empirical solution-quality bounds. According to our results, some of the discovered heuristics substantially improve search efficiency: in the best case, they generated 2.26 times fewer nodes and achieved a 6.73-fold speedup compared to the baseline configurations.

What makes this project especially distinctive is its international, student-led nature. The research was initiated by Xiaojia Li, a student at Carleton College, under the supervision of HSDSLab researchers, as part of the AIT-Budapest Research Abroad Program. It is a great example of how an international research experience can grow into a genuine research collaboration and a published scientific result, and of how international cooperation, student research, and AI research can come together in unexpected and inspiring places.

Congratulations to all co-authors, Xiaojia Li, Csaba Kiss, Dr. Roland Molontay, and József Pintér, on this outstanding achievement!

The full paper is available here: https://link.springer.com/chapter/10.1007/978-3-032-40774-0_27