Talks & Seminars
Title: Modeling Processes and Events
Prof. Niranjan Balasubramanian, Stony Brook University
Date & Time: January 12, 2018 17:00
Venue: Classroom # SIC 301, 03rd Floor, C Block, Department of Computer Science and Engineering, Kanwal Rekhi (KReSIT) Bldg.
In this talk I will describe two structured knowledge extraction efforts. The first part deals with modeling a subset of scientific knowledge related to processes (e.g., evaporation, combustion etc.). Our knowledge about processes often includes typical expectations of how the process will unfold in a situation -- what undergoes the process, what it turns into, and what enables it etc. Automatically extracting this type of role-based knowledge from individual sentences is error prone. Because we are interested in accumulating knowledge about individual processes, we can leverage consistency expectations to perform cross-sentence extraction. I will describe a cross-sentence inference model that improves extraction performance using this idea. In the second part, I will describe our recent work on learning to represent news events using tensor-based models. Schematic knowledge about events can tell us who participates in news events and what roles they play -- e.g., an arrest event typically involves some crime, a suspect, an arrest, and a charging of the suspect. Prior efforts to automatically extract such knowledge used count-based models which suffer from redundancy and sparsity issues. We developed a neural model that produces continuous representations. The key contribution of this work is in learning event tensors which capture context-specific meanings of event predicates (e.g., throwing a bomb is very different from throwing a ball). I will present some empirical results which show that the learned event representations provide benefits for multiple event-related tasks.
Speaker Profile:
Niranjan Balasubramanian is a Research Assistant Professor in Stony Brook University. He runs the language understanding and reasoning lab (LUNR), where his group works on a range of problems in natural language processing. Prior to joining Stony Brook, Niranjan was a post-doctoral researcher in the Turing Center in the University of Washington, where he worked closely with the Allen Institute for Artificial Intelligence (AI2). He graduated with a PhD in Computer Science from the University of Massachusetts Amherst working with Prof. James Allan. More details are available at http://www3.cs.stonybrook.edu/~niranjan/.
List of Talks


Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback