Junshan Zhang
University of California-Davis, CA·Computer Science
Machines increasingly must learn and decide in settings that keep changing: self-driving cars on busy roads, wireless networks with shifting user demand, and learning systems fed by streaming data. A common thread across these projects is designing machine learning methods, especially reinforcement learning, that can adapt quickly when conditions change.
- Fields
Computer Science
Engineering
Mathematics & Statistics
- Topics
reinforcement learning
meta-learning
Actor-Critic
- Grants
- 3 known grants, through Aug 2027
What they work on
- Build CarDreamer, an open-source ecosystem of modular tools for world-model-based self-supervised learning in autonomous driving, through community scoping and planning with academic and industry users.
- Develop the algorithmic and analytical foundations of online bilevel optimization, which solves nested problems on streaming data, so algorithms run faster, scale better and have performance guarantees.
- Design online robust adversarial training, which protects models against misleading inputs, using online compositional bilevel optimization, and accelerate multi-agent meta-learning via online nonconvex bilevel optimization.
- Train deadline-aware wireless scheduling policies with physics-aided offline reinforcement learning, initializing an Actor-Critic learner by behavioral cloning and fixing its overestimation bias and high variance.
- Use meta-reinforcement learning to adapt scheduling policies to changing network conditions, and use reinforcement learning to optimize joint communication and caching decisions for wireless edge caching.
More about the research
Reinforcement learning means learning good actions by trial and feedback. The work spans open-source tools for autonomous driving, mathematical optimization theory, and next-generation wireless networks.
Why it matters
These projects could help make self-driving cars, augmented reality, smart health and smart city systems more dependable. They target faster wireless delivery with very low delays and AI methods that adapt reliably as conditions change. The CarDreamer ecosystem may give researchers, companies and students shared open tools for autonomous driving AI. The optimization work could make machine learning models more robust to deliberately misleading inputs.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Collaborative Research: An Online Bilevel Optimization Framework for Accelerated Learning in Time-varying Environments
NSF · Sep 2024 – Aug 2027
POSE: Phase I: CarDreamer: Towards an Open-Source Ecosystem for World-Model Based Self-Supervised Learning in Autonomous Driving
NSF · Aug 2025 – Jul 2027
NSF-AoF: CNS Core: Small: Reinforcement Learning for Real-time Wireless Scheduling and Edge Caching: Theory and Algorithm Design
NSF · Oct 2021 – Sep 2026
Closest labs
Zejian Zhou
University of Wyoming, WY·Engineering
Model complex autonomous systems as multi-agent systems in which a disrupted component becomes an irrational agent, either passive (an accidental fault) or active (adversarial, such as a cyber-attack).
Hao Xu
University of Nevada-Reno, NV·Engineering·Early-career
Develop a hierarchical game theory (HGT) for large-scale multi-agent systems, such as power management and transportation networks, that maintains optimal efficiency while keeping computing demands manageable as agent numbers grow.
Dinesh Bharadia
University of California-San Diego, CA·Engineering
Develop real-time reinforcement learning algorithms that schedule cellular radio resources by application priority, using the structure of the optimal policy to learn quickly and at scale.
The school
University of California-Davis: 49 labs on PI Pages, $1.0B a year on research, 647 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.