Xi Wang
CS PhD student @ NYU, advised by Shengjie Wang .
I build generative models that don’t just produce fluent outputs — they respect the formal, physical, and scientific constraints of each domain. My research spans language, biological sequences, and robotics, with a shared focus on generating expressively while staying grounded in reality.
Generative Models Language Biological Sequences Robotic Learning
New York University (Courant) · New York, NY
esche.wang@outlook.com
esche.wang@outlook.com
Research
My work explores incorporating formal, physical, and scientific constraints into generative models across multiple domains — making generation both expressive and grounded.
- Generative Models for LanguageControllable text generation with structural and semantic constraints — from decoding-time guidance to constrained fine-tuning.
- Generative Models for Biological SequencesGenerating DNA, RNA, and protein sequences that satisfy biophysical constraints — folding stability, binding affinity, and functional activity.
- Constrained Generative Models for Robotic LearningLearning robotic policies with safety and physics constraints — integrating environment dynamics into generative planning and control.
Selected Publications
- arXivRooted Absorbed Prefix Trajectory Balance with Submodular Replay for GFlowNet TrainingA new training objective and diversity-promoting replay strategy for GFlowNets to address mode collapse in molecular generation.
- Journal of Cheminformatics Clc-db: an open-source online database of chiral ligands and catalystsAn open-source online database of chiral ligands and catalysts for asymmetric synthesis.
- ICLR DeLTAAtropDiff: Data-Scarce Atropisomer Generation via Multi-Task Pretrained Classifier-Guided DiffusionData-Scarce Atropisomer Generation via Multi-Task Pretrained Classifier-Guided Diffusion
- bioRxivUNI-RNA: UNIVERSAL PRE-TRAINED MODELS REVOLUTIONIZE RNA RESEARCHA large-scaled pre-trained language model achieves SOTA performance on RNA donwstream tasks.
News
2026
2025
02.01 Chiralcat published in Artificial Intelligence Chemistry.
2023