Tong Group

Generative Modeling

Developing state-of-the-art generative models including diffusion models and flow-based approaches with applications to molecular design, language modeling, and scientific computing.

Our group develops state-of-the-art generative models with a focus on diffusion and flow-based approaches. We work on both the theoretical foundations—understanding the geometry of generative processes, designing efficient training objectives, and analyzing sampling dynamics—and practical applications spanning molecular design, language modeling, and scientific computing.

Flow matching, introduced in our foundational work on minibatch optimal transport for flow-based generative models, has become a widely adopted training paradigm for continuous normalizing flows and diffusion models. Our current research pushes the boundaries of flow matching in several directions: topological flow matching for structured generation, stochastic flow maps for few-step sampling, and guided generation via Feynman-Kac correctors.

Key Directions

  • Diffusion Models: Progressive inference-time annealing, score-based models, sampling from complex distributions
  • Flow Matching: Optimal transport-based generative modeling, stochastic and topological flow matching
  • Theory & Discretization: Entropy-aware scheduling, superposition of diffusion models, and understanding copying behavior in distillation
  • Applications: Protein design, crystal structure prediction, molecular generation, and language modeling
diffusion-models flow-models deep-learning

People

Alexander Tong

Alexander Tong

Principal Investigator

Zander Blasingame

Zander Blasingame

Postdoctoral Fellow

Antonio Franca

Antonio Franca

PhD Student

Luka Mucko

Luka Mucko

Visiting Student

Katarina Petrović

Katarina Petrović

PhD Student

Charlie B. Tan

Charlie B. Tan

PhD Student

Danyal Rehman

Danyal Rehman

Visiting Researcher

Fred Zhangzhi Peng

Fred Zhangzhi Peng

PhD Student

Selected Publications

← All research areas