Haitong Ma
Papers
5
Total Citations
48
H-Index
3
About
Haitong Ma is a rising researcher at the forefront of safe reinforcement learning (RL) and Bayesian optimization, with a focus on bridging the gap between theoretical guarantees and real-world deployment. His primary research areas include constrained RL for safety-critical systems, model-based RL with formal certificates, multi-agent Bayesian optimization, and sim-to-real transfer for robotics. Ma’s most impactful contribution is the **Feasible Actor-Critic** (2021, 19 citations), which introduced a novel approach to ensuring statewise safety—a significant improvement over expectation-based constraints that can leave individual states unsafe. He further advanced the field with an **uncertainty-aware reachability certificate** for model-based RL (2023, 14 citations), enabling safer training by reducing violations during learning. In multi-agent settings, Ma developed **Gaussian Max-Value Entropy Search** (2023, 10 citations), a sample-efficient algorithm for cooperative black-box optimization. His recent work on **skill transfer and discovery** (2024) leverages spectral decomposition of Markov decision processes to enable robust sim-to-real learning in robotics. With a growing citation record and a clear trajectory toward safety-assured autonomous systems, Ma is establishing himself as a key contributor to trustworthy AI and robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3Gaussian Max-Value Entropy Search for Multi-Agent Bayesian Optimization10 citations · 2023
- 4
- 5