The paper “Unsupervised Multi-kernel Learning for Automated Algorithm Selection” was presented today at PPSN 2026.

Led by Yihang Lu, the study was conducted in collaboration with Carola Doerr from Sorbonne University and Tome Eftimov from the Jožef Stefan Institute.

The paper presents an unsupervised approach to automated algorithm selection that combines four landscape representations—ELA, DeepELA, DoE2Vec, and TransOptAS—through multi-kernel clustering. This enables more generalizable and interpretable solver recommendations without using performance labels during the clustering stage.

As the authors were unable to attend the conference in person, the work was presented on their behalf by Luigi Rovito and Diederick Vermetten. We sincerely thank them for their support and for representing our work at the conference.