News

Uncertainty Quantification and Reasoning for Reliable AI

Despite the remarkable progress in machine learning, AI systems still struggle with uncertainty—manifesting as overconfident predictions, unreliable decisions, and vulnerabilities in safety-critical applications. Addressing these challenges is essential for deploying AI responsibly in real-world scenarios. In this talk, I will present my research on uncertainty quantification and reasoning, a foundational approach to making AI systems more reliable, interpretable, and safe. This talk focuses on three key aspects: (1) Uncertainty quantification and decomposition - analyzing Inherent uncertainties derived from different root causes, aiming to enhance the transparency and trustworthiness of AI systems. (2)Uncertainty Reasoning in Machine Learning: – developing a unified framework to assess and manage the uncertainties associated with AI predictions and decisions, improving model calibration, reliability, and risk assessment in critical safety domains. (3) Reliable Large Language Models (LLMs) – addressing reliability gaps in foundation models by exploring novel uncertainty metrics to mitigate issues such as hallucinations, inconsistencies, and decision-making blind spots. Through the development of fundamental methodologies, theoretical insights, and practical applications, this research contributes to the responsible deployment of AI technologies.

Dr. Xujiang Zhao is a Researcher at NEC Laboratories America. He received his Ph.D. in Computer Science Department at The University of Texas at Dallas in 2022. His research focuses on machine learning, NLP, and data mining, especially in Uncertainty Quantification and Reasoning, Large Language Models, Reinforcement Learning, Natural Language Processing, and Graph Neural Networks Dr. Zhao has published his work in top-tier machine learning and data mining conferences, including NeurIPS, ICLR, AAAI, ACL, NAACL, EMNLP and etc.. He has served on technical program committees for several high-impact venues, such as ICML, NeurIPS, ICLR, KDD, ARR, and AAAI. He also organized and chaired multiple workshops on topics of Uncertainty Quantification, Decision Making, and Trustworthy AI at KDD and AAAI.