Code
Selected open-source projects and paper implementations — more code references on the Publications page.
Active Flow Expansion (ActFlow)
2026 · Out-of-distribution generative discovery
Continued pre-training that expands a flow model's generable set beyond its training distribution via verifier-guided active exploration — with statistical guarantees and empirical gains across small molecules, peptides, and proteins.
Flow Density Control (FDC)
NeurIPS 2025 Spotlight · Distributional fine-tuning of flow & diffusion models
A mirror-descent fine-tuning scheme that reduces optimization of arbitrary utilities under arbitrary divergences to a sequence of standard linear fine-tuning problems, going beyond entropy-regularized reward maximization.