Research
Learning Computational Patterns — Arithmetics, Algebra, and More
Computational tools have long expanded the scope of what humans can compute — how far can deep learning push "computation" itself?
Explaining Vision Models via Game-Theoretic Indices
What is a vision model actually looking at when it makes a decision? We use game theory to visualize and analyze how AI models reach their judgments.
Non-Identical Object Matching
A cheetah and a husky. A photo and an illustration. Humans can match objects that aren't identical but still correspond — we're teaching machines to do the same.
Weather Modification for Heavy-Rainfall Mitigation
Improving deep-learning-based weather forecasting, and exploring physically plausible ways to intervene on extreme weather.
Approximate Vanishing Ideals
Recovering the algebraic structure hidden in noisy point data — a research line bridging computer algebra and machine learning that predates, and continues alongside, our Transformer-based work.
Human-imperceptible Perturbations
Why can perturbations too small for humans to notice make deep learning models fail with high confidence? The theory and practice of adversarial perturbations.