Yusu Wang
University of California-San Diego, CA·Computer Science
Generative AI systems now create images, text and even candidate drug molecules, yet it is unclear why they can produce truly new outputs rather than copying their training data. At the same time, AI is moving from software into the physical world: robots, communication networks and computer chips that must decide and act in real time.
- Fields
Computer Science
Mathematics & Statistics
Engineering
- Topics
generative AI
diffusion models
flow-based generative models
- Grants
- 2 known grants, through Oct 2031
What they work on
- Explain why trained flow-based and diffusion generative models generalize better than the theoretically optimal solution predicts, using geometry, ODE analysis, manifold learning and deep learning theory, to enable controlled generation of novel samples.
- Build a general framework for generative modeling of structured data such as graphs, point clouds and sets, taking a metric-space perspective and using optimal transport.
- Contribute to the TILOS institute's research on AI that adapts and generalizes robustly, teams of embodied agents, and AI running efficiently on chip hardware, with applications such as robotics and chip design.
More about the research
Both grants deal with the mathematical and optimization foundations needed for AI systems that are reliable and broadly usable.
Why it matters
A clearer theory of how generative models produce novel outputs could make generative AI safer, for example by reducing the risk of copying private training data. It could also broaden its use on structured data such as molecular graphs for drug design. Institute work on optimization for embedded AI aims to support more reliable systems in transportation, manufacturing, healthcare, energy, robotics and chip design.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Huibin Zhou
Yale University, CT·Mathematics & Statistics
Determine whether the iterative rectified flow algorithm, which turns high-dimensional density estimation into repeated regression, achieves optimal density and transport estimation in just one step of regression.
Alejandro Ribeiro
University of Pennsylvania, PA·Engineering
Develop optimization and statistical foundations for aligning generative models by exploiting that the problem is strongly concave in distribution space even though it is highly nonconvex in the parameters of deep networks and transformers.
Yongxin Chen
Georgia Institute of Technology, GA·Engineering
Develop theory and algorithms for controlling probability distributions of agents whose dynamics are nonlinear in state and non-affine in control, with optimality guarantees under hard deadlines.
The school
University of California-San Diego: 84 labs on PI Pages, $1.9B a year on research, 637 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.