Abstract:
We develop Dirichlet flow matching on the simplex based on mixtures of Dirichlet distributions as probability paths. In this framework, we derive a connection between the mixtures' scores and the flow's vector field that allows for classifier and classifier-free guidance. On DNA sequence generation tasks, we demonstrate superior performance compared to all baselines in distributional metrics and in achieving desired design targets for generated sequences. Finally, we show that our classifier-free guidance approach improves unconditional generation and is effective for generating DNA that satisfies design targets.
Bio:
Bowen and Hannes are MIT PhD students and the union of their advisors are Bonnie Berger, Tommi Jaakkola, and Regina Barzilay. They worked on generative models for docking, protein structure prediction, molecular dynamics, sequence design, and protein design.
