Event
Booz Allen Hamilton Colloquium: Arthur Redfern, Texas Instruments
Friday, February 19, 2021
3:30 p.m.-4:30 p.m.
Online
Kara Stamets
301 405 4471
stametsk@umd.edu
Arthur Redfern
Machine Learning Lab Manager, Texas Instruments
Title: Binary convolutional neural network design, training and implementation
RSVP: go.umd.edu/redfern
Abstract: Binary CNNs offer the potential for reduced complexity, typically at the expense of accuracy. This presentation will provide a universal function approximator proof for binary neural networks, derive a common building block structure from the proof and use the building block to create a CNN that achieves real MobileNet levels of accuracy on the 2012 ImageNet validation set. Associated binary neural network training and implementation strategies will also be discussed.
Bio: Arthur J. Redfern received a B.S. in 1995 from the University of Virginia and a M.S. and Ph.D. in 1996 and 1999, respectively, from the Georgia Institute of Technology, all in electrical engineering. Following his thesis work on nonlinear systems modeled by the Volterra series, Arthur joined Texas Instruments where he currently manages the Machine Learning Lab. His activities at TI have spanned the areas of machine learning (neural network based applications, software and hardware), high performance computing (software), signal processing for analog systems (ADCs, amplifiers, DACs, design optimization, speakers and touch screens) and physical layer communication system design (DSL, DTV and SerDes). In addition to his work at TI, Arthur also teaches a graduate special topics course on deep learning in the UT Dallas CS department. He has over 25 papers published in refereed conferences and journals and has been granted over 25 US patents.