CIIR Talk Series: Yi Zhang (University of California, Santa Cruz), Learning Inner Monologue and Its Application to Multi-Round RAG

Content

Speaker

Yi Zhang (University of California, Santa Cruz)

Title

Learning Inner Monologue and Its Application to Multi-Round RAG

Abstract

In this talk, I will present our research on developing inner monologue as a core mechanism for adaptive reasoning in large language model systems. Our work explores how models can learn to “think to themselves” — to plan, rehearse, reflect, and refine their reasoning through multi-round self-talk, either with auxiliary modules or within their own internal processes.

Building on the foundation of Retrieval-Augmented Generation (RAG), we show how the inner monologue mechanism transforms retrieval into a dynamic, multi-round process, enabling models to continuously search for and integrate new information, evaluate their intermediate understanding and knowledge boundaries, and adapt their retrieval strategies based on context. Together, these studies point toward a new generation of Retrieval-Augmented AI — systems that not only retrieve knowledge, but also think, imagine, predict, and act with it through self-directed inner monologues.

Bio

Dr. Yi Zhang is a Professor of Computer Science and Engineering at the University of California, Santa Cruz, and Director of the Generative AI Center at UCSC. Her research focuses on information retrieval, natural language processing, and artificial general intelligence. She has received ACM SIGIR Test of Time Award, ACM SIGIR Best Paper Award, the National Science Foundation CAREER Award, and the Air Force Young Investigator Award, as well as research grants from Google, IBM, Bosch, eBay, NEC, and Microsoft.

Dr. Zhang currently serves as Executive Committee Vice Chair for ACM SIGIR and will be the General Chair for SIGIR 2027. She has also served as Program Chair for SIGIR and CIKM, and as Associate Editor for the ACM Transactions on Information Systems. In addition to her academic work, she co-founded Rul.ai, a low/no-code conversational AI agent platform recognized on Forbes’ list of America’s Top 50 Most Promising AI Companies. She has also served as a consultant and technical advisor to several startups and global companies, including Alibaba, HP, and Toyota. Dr. Zhang earned her Ph.D. in Language Technologies from the School of Computer Science at Carnegie Mellon University.

About

The CIIR Talk Series is an initiative for researchers and practitioners working on information retrieval and related disciplines to present their work.

Subscribe to mailing list by sending an email to ciir-talks-request [at] cs [dot] umass [dot] edu (ciir-talks-request[at]cs[dot]umass[dot]edu) with "Subscribe" as the email subject (without the quotation marks), or click here for the Zoom link and zamani [at] cs [dot] umass [dot] edu (reach out to Hamed Zamani) for the passcode.

Host

Hamed Zamani