Science

Astronomical Catalog Cross-matching with Machine Learning

Work in progress.

From Neurons to Neutrons: A Case Study in Interpretability

Kitouni, O., Nolte, N., Pérez-Díaz, V. S., Trifinopoulos, S., & Williams, M. Proceedings of the 41st International Conference on Machine Learning, Vienna, Austria. PMLR 235, 2024.

tldr; We argue that mechanistic interpretability can reveal low-dimensional representations in neural networks that align with human knowledge, as shown through a case study on nuclear physics..

Unsupervised Machine Learning for the Classification of Astrophysical X-ray Sources

Pérez-Díaz, V.S., Martínez-Galarza, J. R., Caicedo Dorado, A., & D’Abrusco, R. (2024). Monthly Notices of the Royal Astronomical Society.

tldr; We classify ~8k astrophysical X-ray sources with an unsupervised learning approach.

GitHub. Web playground. Thesis.

NuDyCLR: Nuclear Dynamic Co-Learned Representations

Pérez-Díaz, V. S., & Trifinopoulos, S. (2023). 2023 REYES Proceedings, 3.

tldr; The best task weighting strategy is simply equal weighting (in this specific case).

Cooperation under infrastructure congestion: the problem of El Farol Bar

Andrade-Lotero, E. J., Pérez-Díaz, V.S. (2023). III Meeting of Mathematics and Statistics Applied to the Social and Economic Sciences.

tldr; We study human cooperation with a mathematical model of people going to a bar.