Keynotes

When Documents Talk Back: The Promise of LLMs in Unstructured Data Query Processing

Speaker: Dr. Nick Koudas

Moderator: Dr. Junzhou Huang

Abstract

In an era where unstructured data dominates, the ability to query documents using natural language is both a critical need and a significant technical challenge. This talk explores how large language models (LLMs) can revolutionize unstructured data querying by enabling natural language interfaces, with a focus on achieving accuracy, reliability, and trust in the query process. We argue that a human-in-the-loop approach is essential to ensure delivery of precise results while adapting to user intent and domain-specific requirements

By integrating human guidance at key stages, one can resolve ambiguities, refine outputs, and incorporate expert knowledge, creating a collaborative and iterative workflow. We demonstrate this approach through the example of natural language to SQL translation, where the LLM leverages user input as needed to steer the computation toward correctness. This method not only achieves superior accuracy on standard benchmarks but also fosters trust by involving users in the process

Ultimately, this talk highlights how combining the strengths of LLMs with human oversight can transform unstructured data into a queryable, actionable resource. By balancing automation with human expertise, we can build systems that deliver both precision and user confidence, unlocking the full potential of unstructured data for a wide range of applications.

Nick Koudas

Professor, University of Toronto, Ontario

Dr. Nick Koudas is a distinguished professor in the Department of Computer Science at the University of Toronto. He specializes in data systems, big data analysis, data science, and applied machine learning. His research focuses on developing algorithms and scalable systems for processing and analyzing massive, diverse datasets, with a particular emphasis on incorporating machine learning techniques as primitives in scalable data systems.

Dr. Koudas' groundbreaking work has earned multiple best paper awards at prestigious international conferences, and he was honored as the 2011 Inventor of the Year (1st Prize) by the University of Toronto. Beyond academia, he is a successful entrepreneur and co-founder of several companies, including Sysomos (now part of Marketwire/NASDAQ), Aislelabs (part of Trapeze Group/Constellation Software), and Workorb.

Leveraging Machine Learning and Generative AI for Data Systems

Speaker: Dr. Surajit Chaudhuri

Moderator: Dr. Gautam Das

Abstract

Significant advancements in Machine Learning and Generative AI have resulted in research and emerging technologies aimed at addressing challenges within data systems. In our group, at Microsoft Research, we explored if machine learning can advance query optimization and if machine learning and generative AI can significantly simplify data cleaning and data transformation. In this talk, I will reflect on our journey and draw a few lessons. I will also briefly mention problems in computer systems where my colleagues in MSR are leveraging advances in Machine Learning and Generative AI.

Surajit Chaudhuri

Technical Fellow and Distinguished Scientist, Microsoft Research | Member of National Academy of Engineering

Surajit Chaudhuri is a Technical Fellow at Microsoft. He manages the Data Systems Group at Microsoft Research, Redmond and he is a member of the MSR Redmond leadership team. His research colleagues and he work closely with engineering teams at Microsoft to develop new technology. His research interests include query optimization for data platforms, data cleaning and transformation, and cloud data services. In recent years, he has worked with colleagues on a set of projects to apply machine learning to some of the open challenges in data systems.

Surajit is a recipient of the ACM SIGMOD Edgar F. Codd Innovations Award and ACM SIGMOD Contributions award. He is the co-winner of the 10-year best paper award in VLDB 2007 and IEEE ICDE Influential Paper Award in 2012. He is a co-winner of the best paper awards in IEEE ICDE 2000, VLDB 2008, and VLDB 2023. Surajit is an ACM Fellow and a member of the US National Academy of Engineering.