Research

I study how learning systems acquire useful representations: state descriptions that preserve task-relevant structure, support prediction under perturbation, and expose uncertainty when information is incomplete.

My interests sit at the intersection of:

• Representation learning: how models organize data into features, latent variables, or state spaces that support generalization and downstream reasoning.
• Generative modeling: diffusion models, energy-based models, and related approaches for learning distributions, dynamics, and sampling procedures.
• Probabilistic and cognitive modeling: Bayesian approaches to learning, inference, perception, and abstraction.
• Statistical-mechanical perspectives on learning: ensembles, entropy, free energy, coarse-graining, and order parameters as tools for thinking about effective descriptions and generalization.