Prof. Joseph (Yossi) Keshet
Dept. of Computer Science, Bar-Ilan University
Trustworthy Neural Models for Speech Processing
In the first part of the talk, I will present my recent work on spoken term detection and keyword spotting. I will present a loss function that allows automatically adjusting the detection threshold while still maintaining high accuracy. Then I will discuss our recent efforts on improving computational efficiency and eliminating the use of dynamic programming (or DTW) in the search process.
The second part of my talk will be devoted to my recent work on trustworthy algorithms. I will introduce an approach named Houdini for generating adversarial examples for complex and structured tasks and demonstrate successful attacks on different speech applications such as speech recognition and speaker verification. I will introduce a novel method to speech steganography, which provides a mechanism to hide a secret spoken message within an ordinary public spoken message. I will conclude this part by briefly presenting a new technique for watermarking deep network models in a black-box way. That is a procedure that allows a tracking mechanism to identify a trained model and can be used to prove ownership multiple times without the loss of credibility of the proofs.
Yossi Keshet received his B.Sc. and M.Sc. degrees in Electrical Engineering in 1994 and 2002, respectively, from Tel Aviv University. He received his Ph.D. in Computer Science from The School of Computer Engineering at The Hebrew Jerusalem in 2008 (Ph.D. dissertation award, EU framework programme for research and innovation). From 1995 to 2002, he was a researcher at IDF and won the prestigious Israeli award, "Israel Defense Prize," for his research and development achievements. From 2008 to 2009, he was a post-doctoral researcher at EPFL (Swiss Federal Institute of Technology Lausanne) and IDIAP Research Institute in Switzerland. From 2009 to 2012, he was a Research Assistant Professor at the University of Chicago and TTI-Chicago. Since 2013 he is an Associate Professor at the Department of Computer Science of Bar Ilan University. Prof. Keshet's research interests are in speech and language processing, and deep learning. He has published over 70 papers in journals and top-tier conferences and co-authored a book.
Zoom link: https://technion.zoom.us/j/94690707872
Wed 16 Jun 2021
Start Time: 10:30
End Time: 11:30
Zoom meeting | The Andrew And Erna Viterbi Faculty Of Electrical & Computer Engineering