Abstract:

The problem of learning interpretable rules from structured data has important theoretical and practical ramifications in the fields of machine learning and program synthesis. Datalog, a declarative logic programming language, has emerged as a popular medium for studying this problem due to its rich expressivity and scalable performance. I will present search-based and constraint-solving techniques to learn Datalog programs from relational input-output data. The techniques address previously open problems as well as pose new challenges, spanning data-efficient learning, tolerating noise, supporting expressive features of Datalog, learning without syntactic bias, and scaling to very large search spaces.

Biography:

Mayur Naik is an Associate Professor of Computer and Information Science at the University of Pennsylvania. His research spans programming languages related topics with the overarching goal of making software better, safer, and easier to build and maintain. His current focus concerns developing scalable techniques to reason about programs by combining machine learning and formal methods. He is also interested in foundations and applications of neuro-symbolic approaches that synergistically combine deep learning and symbolic reasoning. He received a Ph.D. in Computer Science from Stanford University in 2008. Previously, he was a researcher at Intel Labs, Berkeley from 2008 to 2011, and a faculty member at Georgia Tech from 2011 to 2016.