Colloquium - Meng Jiang, "Lessons Learned from Enhancing Knowledge and Reasoning for (Large) Language Models"

Abstract: 
During the past two years, my students and I have been designing, developing, and evaluating methods to enhance knowledge and reasoning capabilities of language models. In this talk, I’d like to share the lessons we learned from this experience. The talk will answer where the language models can gain knowledge and where they can learn to use the knowledge effectively from, and how. Some interesting questions are: Can large language models (LLMs) verify themselves when answering hard questions? Can LLMs improve themselves by learning from synthetic data? What can we expect the AI’s knowledge and reasoning capabilities to be in the next couple of years, as the foundational skills behind agents?

Bio:
Meng Jiang is Associate Professor of Computer Science and Engineering at the University of Notre Dame. He directs the Data Mining Lab. His research focuses on mining text and graph data for applications such as user modeling, recommender system, question answering, online education, mental health, and material discovery. He received SIGSOFT distinguished paper award in 2021, NSF CAREER award in 2022, and EMNLP outstanding paper award in 2023. He has delivered 14 tutorials and organized nine workshops. The Knowledge-Augmented NLP workshop series he’s been organizing has expanded from 80 participants in AAAI 2023 to over 300 participants in ACL 2024, and it’s being organized the fourth time at NAACL in May 2025. Jiang has also served on the Organizing Committee for ACM SIGKDD for four years.