[Ml-stat-talks] Princeton Optimization Seminar: Vikas Sindhwani, Google Research, Thursday Oct. 8, 4:30PM
Amir Ali Ahmadi
a_a_a at princeton.edu
Thu Oct 1 19:40:19 EDT 2015
----- Princeton Optimization Seminar -----
DATE: Thursday, October 8, 2015
TIME: 4:30 PM
LOCATION: Sherrerd Hall 101
SPEAKER: Vikas Sindhwani, Google Research
TITLE: Structured Transforms for Small-footprint Deep Learning<https://orfe.princeton.edu/abstracts/optimization-seminar/structured-transforms-small-footprint-deep-learning>
We consider the task of building compact machine learning pipelines suitable for deployment on storage and power constrained devices such as smartphones, wearable computers or mobile robots. We propose a unified framework to learn a family of structured parameter matrices that are characterized by the notion of low displacement rank. Our structured transforms admit fast function and gradient evaluation, and span a rich range of parameter sharing configurations whose statistical modeling capacity can be explicitly tuned along a continuum from structured to unstructured. Experimental results show that these transforms can impressively accelerate test-time inference and forward/ backward passes during training; and offer superior accuracy-compactness-speed tradeoffs in comparison to a number of existing model compression techniques. In keyword spotting applications in mobile speech recognition, our methods are much more effective than standard linear low-rank bottleneck layers and nearly retain the performance of state of the art models, while providing more than 3.5-fold compression.
Vikas Sindhwani is staff research scientist at Google, NY. Previously, he led the Machine Learning Group at IBM T. J. Watson Research Center. His research interests span non-parametric and high-dimensional learning, distributed numerical optimization, and applications in perceptual Artificial Intelligence problems. He won the best paper award at UAI 2013, and the IBM Pat Goldberg Memorial award in 2014. He has a PhD degree in Computer Science from the University of Chicago, and a Bachelors degree in Engineering Physics from the Indian Institute of Technology, Bombay. His publications are available at: http://vikas.sindhwani.org.
(If you want to meet Vikas on Thursday afternoon, please email me [ a_a_a at princeton.edu ].)
*** If you would like to subscribe to the mailing list of the optimization seminar series, please visit the link below or send an email with
SUBSCRIBE opt-seminar in the body to listserv at lists.princeton.edu<mailto:listserv at lists.princeton.edu>.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ml-stat-talks