Shmuel Ur is an Israeli computer scientist and researcher in artificial intelligence, known for his work on testing and verification of machine learning systems. His career includes significant contributions at IBM Research and collaborations with academic institutions, focusing on ensuring reliability and safety in AI applications.
Ur's early work centered on software testing and formal verification, areas in which he published extensively during his tenure at IBM. He later shifted his focus to the challenges posed by machine learning, particularly the need for robust testing methodologies for neural networks and other AI models. His research addresses the gap between traditional software testing and the unique properties of learned systems, such as their non-determinism and sensitivity to input perturbations.
Early Career and IBM Research
Ur joined IBM Research in the 1990s, where he worked on hardware and software verification. At IBM's Haifa Research Laboratory in Israel, he contributed to projects involving model checking and test generation for complex systems. His work during this period laid the groundwork for his later interest in AI verification, as he applied formal methods to increasingly sophisticated software systems.
Transition to AI Verification
By the mid-2010s, Ur had turned his attention to Machine learning and Deep learning systems. He recognized that traditional testing techniques were insufficient for validating the behavior of Neural network models, which often operate as black boxes. His research proposed methods for systematically generating test cases that expose vulnerabilities in AI systems, including adversarial examples that cause misclassification. This work aligns with broader efforts in the AI community to ensure the robustness of models used in critical applications.
Academic and Industry Collaborations
Ur has held visiting positions and collaborated with researchers at institutions such as the Technion – Israel Institute of Technology and Tel Aviv University. He has also worked with industry partners, including Samsung Research and Intel, on projects related to AI safety. His consulting engagements have focused on developing verification frameworks for AI products, though specific details of these collaborations are not publicly documented.
Key Publications and Contributions
Ur has authored or co-authored over 50 peer-reviewed papers, many in venues such as the International Conference on Software Engineering and the IEEE Transactions on Software Engineering. Notable works include studies on coverage criteria for neural networks and methods for detecting bias in AI models. His papers are frequently cited by researchers working on AI testing, indicating his influence in this niche field.
Later Work and Current Focus
In recent years, Ur has explored the intersection of AI and formal verification, advocating for a more rigorous approach to validating AI systems before deployment. He has spoken at conferences about the ethical implications of unreliable AI, emphasizing the need for transparency and accountability. As of 2025, he continues to contribute to the field through independent research and advisory roles.
Legacy and Impact
Ur's work has helped establish testing and verification as a critical subfield within Artificial intelligence research. By adapting classical verification techniques to the challenges of machine learning, he has influenced both academic research and industry practices. His efforts are part of a broader movement to ensure that AI systems are not only powerful but also trustworthy.
References
This article is based on publicly available information, including Ur's publication record and conference presentations. Specific dates and affiliations are drawn from his professional history, though some details remain unverified due to limited public sources.