Runqian (Ray) Wang

I am a PhD student at UC Berkeley, advised by Prof. Alexei Efros. I graduated from MIT with double major in AI and Math. I am fortunate to have worked with Prof. Kaiming He, Prof. Yilun Du, and Dr. Zhirong Wu. I also interned at Microsoft Research Asia and MIT-IBM Watson AI Lab.

I hope to work on general problems in deep learning through the lens of computer vision. I am currently interested in generative models and representation learning.

Email  /  CV  /  Github  /  Google Scholar

For a full list of publications, see Google Scholar

Plug-and-play representation regularizer for generative modeling that brings consistent improvement across different settings. Follow-up works also show promising adoptation in robotics and NLP.

New generative model that directly learns equilibrium dynamics using an implicit energy-based formulation. Exceeds Flow Matching in performance, supports optimization-based sampling, and naturally performs multiple downstream tasks.

Equilibrium Forcing improves and extends Equilibrium Matching to video generation, achieving better video generation quality and sampling speed through adaptive sampling.

Reframes ARC-AGI dataset as an image-to-image translation problem, achieving competitive performance with those of leading LLMs.

Enables nearly data-free and compute-efficient transfer of existing PEFT modules trained on old base model to new base models, while at least preserving, in most cases improve, performance.