- This event has passed.
CDS Seminar(23 August, 2019): Modeling Event Knowledge
Fri, 23 Aug 19 @ 11:00 AM -- 12:00 PM
Department of Computational and Data Sciences
SPEAKER : Niranjan Balasubramanian at Stony Brook University
TITLE : “Modeling Event Knowledge ”
Date/Time : August 23,2019 (Friday) 11:00 AM
Venue : #102, CDS Seminar Hall.
Knowledge about how situations unfold can be helpful for reasoning about descriptions of real-world events. In this three part talk, I will first present tensor-based composition models for learning representations of events from predicate-argument triples. When learning from these reduced representations, even a small change to the surface form can result in a huge shift in meaning. We address this challenge by modeling predicates as tensors. I will describe how we can scalably learn these tensors from unlabeled text. In the second part of the talk, I will discuss our recent work on modeling variations in script-like knowledge. There are many variations to any particular type of scenario — for example, a protest scenario can be peaceful or violent. I will present a variational autoencoding model that is designed to handle the hierarchical nature of script like knowledge. In the third part, I will briefly mention other on-going work in question answering, language generation, and efficient NLP for mobile devices.
Niranjan Balasubramanian is an Assistant Professor in the Computer Science department at Stony Brook University. Prior to joining Stony Brook, he was a post-doctoral researcher in the University of Washington, and one of the early members of the Allen Institute for Artificial Intelligence. Niranjan has a PhD in Computer Science from the University of Massachusetts Amherst. He likes food and wine so much that he sometimes mentions about it in his bio.
Host Faculty: Prof. Partha Pratim Talikdar