Ziqing Lu, left, and Weiyu Xu, right, pose for a photo at the Seamans Center for the Engineering Arts and Sciences

Ziqing Lu honored with Best Thesis Award for research on building more reliable reinforcement learning models

Thursday, September 10, 2026

Ziqing Lu has received the Best Thesis Award from the University of Iowa's Program of Applied Mathematics and Computational Science (AMCS) for her dissertation, “Adversarial Robustness of Reinforcement Learning Systems.” Lu conducted her research under the mentorship of Weiyu Xu, professor of electrical and computer engineering.

Ensuring artificial intelligence systems operate reliably remains a significant challenge. For example, changing just a few pixels in an image of a cat may cause a neural network to identify it as a dog, even though it still looks like a cat to a human observer. Likewise, AI systems can struggle with tasks that humans find straightforward, such as identifying traffic lights in CAPTCHA images. These kinds of vulnerabilities were central to Lu's research.

"Overall, my dissertation aimed to improve our understanding of when and why reinforcement learning systems and neural networks are vulnerable, and to support the development of more reliable AI systems," Lu said.

Lu's dissertation focused on reinforcement learning, a branch of AI in which agents learn by interacting with their environment. While reinforcement learning has applications in robotics, autonomous systems, and large language models, these systems can be vulnerable to small changes in their observations or surroundings. Such weaknesses are especially concerning in safety- and security-critical applications, including autonomous vehicles, robotics, and security and defense systems.

"I have been curious as to why this phenomenon happens [when neural networks get fooled], so I'm quite excited about Ziqing's findings," Xu said. “It’s an important step in understanding their vulnerabilities.”

Lu has published her PhD work as first author at top AI conferences including the AAAI Conference on Artificial Intelligence, International Conference on Learning Representations, International Conference on Machine Learning, and also in the IEEE Signal Processing Society flagship journal, IEEE Transactions on Signal Processing.

Now a postdoctoral researcher in the Department of Computer Science at Wake Forest University, Lu continues to study reinforcement learning and develop frameworks for safer, more robust AI systems. Her work aims to improve the reliability and stability of AI agents operating in real-world environments when they face unexpected perturbations or adversarial attacks.