Bio
Jay Pujara is a research assistant professor of Computer Science at the University of Southern California as well as research team lead and director of the Center on Knowledge Graphs at USC’s Information Sciences Institute. His research focuses on artificial intelligence, specifically knowledge graphs and statistical relational learning. Jay is the author of over fifty peer-reviewed publications, has received four best paper awards, and been featured in AI Magazine. For more information, visit https://www.jaypujara.org
Talk Title: Building Robust Systems with Structured and Implicit Knowledge
Abstract: Financial applications are often driven by complex workflows, a strong dependency on quantitative knowledge, and implicit knowledge about how concepts relate. In this talk, I will describe our work on building robust AI systems that tackle these challenges in two specific domains: conversational agents with common sense and table understanding systems that leverage structural relationships. Common sense is an essential ingredient for understanding the world and others, but language models have many deficits in using common sense. Our research has quantified these deficits, built systems that integrate common sense into generative modeling, and developed models to actively select the most relevant implicit knowledge to achieve goals. However, in financial application, the relevant knowledge goes beyond common sense, and relies on structured quantitative datasets. I will discuss our work on the task of table understanding – translating complicated tabular datasets into usable knowledge. I will present work on understanding values, regions, and relationships in tables, matching concepts in text and tables, and fact verification and question answering using structured data.