Joseph (Yossi) Keshet

Department of Computer Science
Bar Ilan University,
Room 44, Bldg 109
Ramat Gan, 52900, Israel

Email: jkeshet at

Tel: +972-3-738-4378

I am a faculty member in the Department of Computer Science at Bar-Ilan University. My research interests concern both machine learning and computational study of human speech and language. In machine learning my research is focused on deep learning and structured prediction, while my research on speech and language is focused on speech processing, speech recognition, acoustic phonetics, and pathological speech.



Copyright disclaimer

This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.

Resources & Code

Open Source Projects


Autovot is a software package for automatic measurement of voice onset time (VOT), using an algorithm which is trained to mimic VOT measurement by human annotators.


StructED is a collection of machine learning algorithms for structured prediction. Structured tasks are distinctive: each task has a unique evaluation metric, its own set of feature functions, and in many cases a non-standard inference. Consequently, implementation of a machine learning system that utilizes structured prediction is complex and involved. StructED handles this inherent complexity by introducing specially-designed interfaces for an evaluation function, a set of feature functions, and a decoder. The collection of all the training algorithms is implemented using the same interfaces, and adding a new training algorithm is straightforward. The library is written in Java, hence is platform independent.

A GPU-Tailored Approach for Training Kernelized SVMs

This is an implementation of an algorithm for efficiently training a binary or multiclass kernelized Support Vector Machine (SVM) on a Graphics Processing Unit (GPU). Our approach is distinguished from earlier work in that it cleanly and efficiently handles sparse datasets, and is also specifically designed to take maximal advantage of the graphics hardware. This leads to different algorithmic choices then those preferred in serial implementations, whereas previous GPU implementations have tended to be straightforward parallelizations of successful serial algorithms. Our library is orders of magnitude faster then existing CPU libraries, and up to several times faster than prior GPU approaches. We have made every effort to ensure that this library is not only efficient, but also very easy-to-use.