Research ThemesOur research explores the geometry of data and models across multiple areas. Graph Neural NetworksUnderstanding how GNNs generalize to unseen nodes, through optimal transport and geometric perspectives, and improving node classification.
Trustworthy AIDeveloping methods for machine unlearning, privacy-preserving ML, and robust evaluation of AI systems.
ML for ScienceApplying machine learning to scientific discovery, including drug design, molecular modeling, and computational biology.
Language ModelsUnderstanding and building reliable LLM agents: how harness design interacts with post-training, multi-turn collaboration with experts, energy-efficient multi-agent systems, and multimodal reasoning.
Generative ModelsUnderstanding the learning dynamics of diffusion models, composing pre-trained generators for multi-objective generation, and generating structured data such as graphs.
|