Kaidi Xu

Kaidi Xu is an Assistant Professor in the Department of Computer Science at Drexel University's College of Computing & Informatics. His research focuses on Trustworthy AI, with expertise in formal verification of neural networks, adversarial attacks (especially in the physical world), and certified defenses. He actively publishes in top-tier conferences including NeurIPS, ICML, ICLR, CVPR, and AAAI, and leads the award-winning research team 'alpha-beta-crown'.

• PhD in Computer Science, Northeastern University (2021)
• MS in Computer Science, University of Florida (2017)
• BS in Computer Science, Sichuan University (2015)

Dr. Xu's research spans critical areas in AI security and robustness. He investigates formal methods to verify neural network behavior, develops techniques to defend against real-world adversarial manipulations (such as the famous 'Adversarial T-shirt'), and explores model compression and explainability. His work bridges theoretical guarantees with practical applications in healthcare, material science, and autonomous systems.

His recent publications reflect a strong trend toward certified robustness, interdisciplinary applications, and formal verification across vision, language, and multimodal systems. He has consistently published at NeurIPS, ICML, CVPR, and ACL, demonstrating sustained impact in both machine learning and computer vision communities.

• Winner of VNN-COMP'21 with highest score
• Three-time VNN-COMP champion (2021–2023) with team alpha-beta-crown
• Faculty Research Excellence Award, CCI@Drexel (2024)
• Recipient of multiple Carleone Faculty Awards (2025)
• NSF grant recipient for projects on transit systems and material synthesis

Dr. Xu advises PhD students, including Jinhao, and has secured significant external and internal funding, including multiple NSF grants and Drexel internal awards. He is actively recruiting motivated students with strong machine learning backgrounds. He also contributes to the academic community as an Area Chair for NeurIPS 2025, organizer of workshops like GenAI4Health@AAAI 2025, and frequent program committee member.

He leads the 'alpha-beta-crown' research team, known for its leadership in neural network verification and repeated success in the VNN-COMP competitions. The team focuses on developing scalable, sound, and complete verification tools for deep learning models, pushing the frontier of AI safety and reliability.