My research focuses on principled multimodal generative AI. I develop models that connect understanding of the world with generation, with an emphasis on controllability and generalization.
I received my Ph.D. from Carnegie Mellon University in August 2026, advised by Prof. Kun Zhang and Prof. Peter Spirtes. Previously, I interned at Adobe Research and Google.
ACL 2026 · Main
We improve reasoning in diffusion LLMs by adding process rewards to GRPO-based reinforcement learning (RL) during post-training. These rewards compare answer success rates before and after a denoising interval, giving the model credit for progress toward a correct answer.
No human-labeled reasoning steps or separate reward model required.
CVPR 2023 · Highlight
Integrated into Adobe Photoshop and Firefly.
Multi-turn image editing with text and shape guidance: add a hat, then sunglasses, then a blue shirt—building on each previous edit.
ICLR ’26 & ’27, NeurIPS ’26
ICDM ’24
ICML, NeurIPS, ICLR, CVPR, ICCV, ECCV, UAI, AISTATS, SIGGRAPH
TPAMI, TIP, AI, CSUR, TNNLS, PR, TMM, JASA, TVCG, IJCV
