Constrained Diffusion for Protein Design with Hard Structural Constraints
A constrained diffusion approach for incorporating hard structural requirements into protein design.
Read paper
About Me
I am a Ph.D. candidate in Computer Science at the University of Virginia, advised by Dr. Ferdinando Fioretto. My research develops generative models that satisfy hard constraints arising in scientific and engineering applications. I combine diffusion and flow-based models with mathematical optimization to improve the reliability, controllability, and efficiency of generative AI. My work spans protein and molecular design, multi-robot planning, physical systems, and accelerated language generation.
Research Interests
- Constrained generative modeling Developing diffusion and flow-based models that satisfy discrete, structural, and physical constraints.
- AI for science and engineering Applying constraint-aware generation to protein design, molecular discovery, physical systems, and robotics.
- Speculative decoding with dLLMs Designing diffusion language models for faster speculative decoding with parallel generation.
Selected Highlights
Spotlight
NeurIPS 2025
Training-Free Constrained Generation with Stable Diffusion Models
Oral
MLSys 2026
SpecDiff-2: Scaling Diffusion Drafter Alignment for Faster Speculative Decoding
Award
DARPA Disruptive Idea Award
NeuS 2025
Best Student Paper
AI4D3 at NeurIPS 2025
Constrained Molecular Generation with Discrete Diffusion for Drug Discovery
Featured Research
Training-Free Constrained Generation with Stable Diffusion Models
A training-free method for guiding Stable Diffusion models toward outputs that satisfy specified constraints.
SpecDiff-2: Scaling Diffusion Drafter Alignment for Faster Speculative Decoding
A diffusion-based drafting approach designed to accelerate speculative decoding for language generation.
Read paperLatest Updates
January 2026
Two papers accepted at ICLR 2026 and MLSys 2026; SpecDiff-2 was selected for an oral presentation.
December 2025
Constrained Molecular Generation with Discrete Diffusion for Drug Discovery received the Best Student Paper award at the AI4D3 Workshop at NeurIPS 2025.
September 2025
Two papers accepted at NeurIPS 2025: Training-Free Constrained Generation with Stable Diffusion Models, selected as a Spotlight, and Constrained Discrete Diffusion.
May 2025
Simultaneous Multi-Robot Motion Planning with Projected Diffusion Models was accepted at ICML 2025, and our neuro-symbolic diffusion work received the DARPA Disruptive Idea Award at NeuS 2025.
April 2025
Speculative Diffusion Decoding: Accelerating Language Generation through Diffusion was presented as an oral at NAACL 2025.