Mingyan Liu

Papers

Total Citations

H-Index

About

Mingyan Liu is a leading researcher in the security and robustness of machine learning systems, with a particular focus on deep reinforcement learning (DRL) and adversarial machine learning. Her most-cited work, "Characterizing Attacks on Deep Reinforcement Learning" (2019, 52 citations), provides a foundational taxonomy of adversarial threats to DRL models, systematically analyzing how small perturbations to observations can compromise agent performance. This paper critically evaluates the practicality of existing attack methods, highlighting limitations such as the unrealistic assumption of full model access and prohibitive computational costs. By exposing these vulnerabilities, Liu has helped shape the field's understanding of security gaps in autonomous decision-making systems. Her contributions extend to developing more realistic threat models and robust training methods, influencing both academic research and practical deployment of AI systems. With her work bridging the gap between theoretical security analysis and real-world applicability, Liu continues to drive progress in making machine learning systems safer and more trustworthy for critical applications.