黑料福利网

Topic 

Building Next Generation of Trustworthy AI Systems

Abstract

As AI systems move from isolated prediction tasks to autonomous, tool-using, multi-agent workflows, trustworthiness must be built into the system architecture rather than evaluated only after deployment. This talk presents a framework for next-generation trustworthy AI that combines agentic reasoning with formal modeling, verification, and execution-time guardrails. Motivated by the challenge of executing complex, multi-step Standard Operating Procedures, the work introduces SpecMass, a multi-agent system that translates natural-language SOPs into verifiable formal models and uses them to guide downstream task execution. Its 鈥淧lan and Evolve鈥� paradigm allows agents to generate, verify, revise, and execute plans through a closed-loop process, where model checking, symbolic feedback, and structured orchestration reduce hallucinations, invalid transitions, and procedural failures. By grounding agent behavior in formal artifacts such as NuSMV models and Kripke-structure-based automata, the approach moves beyond prompt-based reasoning toward AI systems that are auditable, adaptive, and mathematically constrained. The broader vision is to enable trustworthy AI systems that can operate reliably in high-stakes domains such as engineering, enterprise automation, healthcare, finance, and cyber-physical systems where correctness, compliance, and safety are essential.

About the speaker

Dr. Apurva Narayan is an Associate Professor in the Departments of Computer Science and Electrical & Computer Engineering at the University of Western Ontario, where he leads the Intelligent Data Science Lab (IDSL). He also holds affiliated and adjunct academic appointments with the University of British Columbia and the University of Waterloo and supervises a cross-institutional research team spanning these universities.

Dr. Narayan is an established researcher whose work advances the next generation of trustworthy, dependable, and verifiable AI systems. His research spans machine learning, data science, software engineering, decision-making under uncertainty, robust and explainable AI, formal methods, multi-agent AI, runtime monitoring, and neuro-symbolic reasoning. His work has been published in top-tier journals and computer science conferences, with applications across Industry 4.0, cyber-physical systems, AI safety and security, trustworthy AI, chemoinformatics, and bioinformatics.

His research program has attracted approximately CAD $11 million in funding, with support from NSERC, NFRF, Mitacs, NSF, and competitive research awards from Amazon, Microsoft, and Meta. Dr. Narayan serves on the program committees of several leading venues, including ECML, AAAI, NeurIPS,  IJCAI, QRS, ASE, FSE, PKDD, KDD, EMSOFT and many more. His work has received best paper awards and nominations at multiple venues, and he is a recipient of the Young Systems Scientist Award from the Systems Society of India and Western University鈥檚 2026 Florence Bucke Faculty of Science Prize.

Your host

, Assistant Professor in the Data Mining group of the department of Mathematics and Computer Science at the Eindhoven University of Technology (黑料福利网).

Registration is required but free of charge.