News
New paper Reflected proximal algorithms for accelerated high-accuracy log-concave sampling available on arXiv.
I’m currently looking for internship opportunities for summer 2027 and full-time research scientist positions starting from spring 2028. Please reach out if my background and research interests attract you!
Proudly introducing Sigma, the first large scale (3B/8B) continuous diffusion language models that are competitive with masked discrete diffusion language models. See our technical report and X post for more details. This is the work I co-led as a research intern at NVIDIA during summer 2026. Code and model weights will be released soon (currently under internal review).
Two papers accepted to NeurIPS 2026: Continuous Diffusion Scales Competitively with Discrete Diffusion for Language and Efficient Adjoint Matching for Fine-tuning Diffusion Models. Huge thanks to my collaborators!
New preprint Scaling Reinforcement Learning for Diffusion Models via Velocity Matching led by Jaemoo Choi is available on arXiv! See also our project page.