About this document
Fisher Flow Matching for Discrete Data by aref.shahbakhsh1998 is a document available to read on EtoBox.
The document introduces F ISHER -F LOW, a novel flow-matching model for generative modeling over discrete data, leveraging the Fisher-Rao metric to treat categorical distributions as points on a statistical manifold. This approach allows for continuous reparameterization of discrete data, enabling improved training dynamics and performance in tasks such as DNA sequence design. Empirical evaluations demonstrate that F ISHER -F LOW outperforms existing diffusion and flow-matching models on various benchmarks.
- Author
- aref.shahbakhsh1998
- Language
- EN