About
I'm Jing Yuan (袁璟), based in Shanghai. I was most recently a Research Scientist at ByteDance Seed, on the AI for Science team led by Prof. Quanquan Gu, where my work centered on generative modeling and cryo-electron microscopy.
Before that, I worked on applied machine learning in ByteDance's Education team, and on computer vision and medical imaging at SenseTime. I earned my M.S. from the University of Science and Technology of China in 2019, working on image saliency detection.
These days, I'm spending some time on mathematics, mostly as a way to step outside my usual toolkit and look at familiar problems from a different angle. I'm also using the time to read, experiment, and think more freely about intelligence and learning.
Research Taste
I have a bias toward taking things apart before adding anything new: what is actually broken, what is fundamental to the problem, what is just baggage from the current approach and what is the smallest change that would make a real difference? I like experiments where an idea has somewhere to die — small, controlled settings where you can tell why something worked, or why it didn't. If the signal survives there, then scale it. If it only appears after enough knobs, compute, and storytelling, I'm usually less interested. The best result, to me, is one that changes how I think the system should be built, not just a better number in a table.
currently interested in: continual adaptation, generative models, and whatever eventually makes AGI.dev compile