Karen Livescu, PhD, Toyota Technical Institute of Chicago – MIDAS/AI Seminar Series

Karen Livescu

Associate Professor

Toyota Technical Institute of Chicago

“(How) Should We Use Domain Knowledge in the Era of Deep Learning? 

(A Perspective from Speech Processing)”

Abstract:  Deep neural networks are the new default machine learning approach in many domains, such as computer vision, speech processing, and natural language processing.  Given sufficient data for a target task, end-to-end models can be learned with fairly simple, almost universal algorithms.  Such models learn their own internal representations, which in many cases appear to be similar to human-engineered ones.  This may lead us to wonder whether domain-specific techniques or domain knowledge are needed at all.

This talk will provide a perspective on these issues from the domain of speech processing.  It will discuss when and how domain knowledge can be helpful, and describe two lines of work attempting to take advantage of such knowledge without compromising the benefits of deep learning.  The main application will be speech recognition, but the techniques discussed are general.

Bio: Karen Livescu is an Associate Professor at TTI-Chicago.  She completed her PhD and post-doc in electrical engineering and computer science at MIT and her Bachelor’s degree in Physics at Princeton University.  Karen’s main research interests are at the intersection of speech and language processing and machine learning.  Her recent work includes multi-view representation learning, segmental neural models, acoustic word embeddings, and automatic sign language recognition.  She is a member of the IEEE Spoken Language Technical Committee, an associate editor for IEEE Transactions on Audio, Speech, and Language Processing, and a technical co-chair of ASRU 2015 and 2017.

This seminar is co-sponsored by Aritificial Intelligence, Computer Science and Engineering.

For more information on MIDAS or the Seminar Series, please contact midas-contact@umich.edu. MIDAS gratefully acknowledges Northrop Grumman Corporation for its generous support of the MIDAS Seminar Series.