Yuxin Chen
University of Pennsylvania, PA·Mathematics & Statistics
Many modern data problems involve learning a huge number of unknowns from data that are scarce, noisy, or costly to collect. Examples include an algorithm that learns by trial and error in a clinical trial, or an imaging method that must reconstruct a picture from incomplete measurements.
- Fields
Mathematics & Statistics
Computer Science
Engineering
- Topics
reinforcement learning
sample complexity
function approximation
- Grants
- 2 known grants, through Sep 2027
What they work on
- Characterize the fundamental limits on sample and computational complexity of reinforcement learning, meaning how much data and computing an agent needs to learn good decisions in an unknown environment.
- Design computationally efficient reinforcement learning algorithms that use function approximation and are provably near-optimal in sample use, for both single-agent and multi-agent settings.
- Develop performance guarantees and uncertainty quantification schemes for RL algorithms, using finite-sample analysis tools that handle the statistical dependencies in sequential decision data.
- Construct optimal confidence intervals for continuous parameters in nonconvex low-complexity models, using new nonconvex estimation and de-biasing methods.
- Build fast nonconvex algorithms and uncertainty assessment procedures for unknown discrete variables, and unify convex relaxation and nonconvex optimization in one uncertainty framework tested on applications such as shape matching and water-fat imaging.
More about the research
This research builds the mathematical theory behind such methods. It asks how much data and computation they really need, and how confident we can be in their answers. One project studies reinforcement learning, and the other studies how to measure uncertainty in large-scale estimation.
Why it matters
Algorithms that learn from limited data are increasingly used where mistakes are costly, such as clinical trials, autonomous systems, and medical imaging. Clear limits on how much data these methods need, together with valid error bars on their outputs, could help scientists, engineers, and clinicians judge when to trust an algorithm's decisions. This work could also support safer use of reinforcement learning in settings where data are scarce.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Chi Jin
Princeton University, NJ·Computer Science·Early-career
Identify the general structures of function classes, such as those represented by neural networks, under which model-free reinforcement learning is tractable, and design provably efficient algorithms that also handle model misspecification.
Pan Xu
Duke University, NC·Computer Science·Early-career
Develop distributionally robust reinforcement learning methods that use minimax optimization over a set of possible environment dynamics, so that policies trained in a simulator perform reliably in an unknown target domain, with finite-sample guarantees.
Navid Azizan Ruhi
Massachusetts Institute of Technology, MA·Engineering·Early-career
Develop Hard-Constrained Neural Networks (HardNet), which use a differentiable projection layer to guarantee input-dependent constraints, and embed them in reinforcement learning to produce certifiably safe policies.
The school
University of Pennsylvania: 90 labs on PI Pages, $2.2B a year on research, 750 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.