Language Technologies Institute Colloquium

  • NOAH SMITH
  • Professor of Computer Science &Engineering
  • Paul G. Allen School of Computer Science & Engineering, University of Washington
  • Senior Research Manager, Allen Institue for Artificial Intelligence
Colloquium

Slouching Towards Understanding: Rational Recurrences and Neighbor-Based Neural Networks

Despite their often-discussed advantages, deep learning methods largely disregard theories of both learning and language.  This makes their prediction behavior hard to understand and explain.  In this talk, I will present two ideas to improve "deep" NLP models in this regard, without sacrificing accuracy.  The first, rational recurrences, comprise a family of RNNs that obey a particular set of rules about how to calculate hidden states, and hence correspond to parallelized weighted finite-state pattern matching.  Many recently introduced RNNs turn out to be members of this family, and the WFSA view lets us derive some new ones.  The second idea turns to classification, replacing the final "softmax" in any neural classifier with a weighted average of training instances.  This approach offers transparent explanations in terms of neighbors and distances.

Noah Smith is a Professor in the Paul G. Allen School of Computer Science & Engineering at the University of Washington, as well as a Senior Research Manager at the Allen Institute for Artificial Intelligence. Previously, he was an Associate Professor of Language Technologies and Machine Learning in the School of Computer Science at Carnegie Mellon University. He received his Ph.D. in Computer Science from Johns Hopkins University in 2006 and his B.S. in Computer Science and B.A. in Linguistics from the University of Maryland in 2001.

His research interests include statistical natural language processing, machine learning, and applications of natural language processing, especially to the social sciences. His book, Linguistic Structure Prediction, covers many of these topics. He has served on the editorial boards of the journals Computational Linguistics (2009–2011), Journal of Artificial Intelligence Research (2011–present), and Transactions of the Association for Computational Linguistics (2012–present), as the secretary-treasurer of SIGDAT (2012–2015 and 2018–present), and as program co-chair of ACL 2016. Alumni of his research group, Noah's ARK, are international leaders in NLP in academia and industry; in 2017 UW's Sounding Board team won the inaugural Amazon Alexa Prize. Smith's work has been recognized with a UW Innovation award (2016–2018), a Finmeccanica career development chair at CMU (2011–2014), an NSF CAREER award (2011–2016), a Hertz Foundation graduate fellowship (2001–2006), numerous best paper nominations and awards, and coverage by NPR, BBC, CBC, New York TimesWashington Post, and Time.  

Faculty Host: Yulia Tsvetkov

Refreshments at 4:00 pm - 5th Floor LTI

For More Information, Please Contact: 
Keywords: