~~(The talk will be given in English)

Speaker:   Dr. Ofer Meshi
                       Toyota Technological Institute at Chicago

Monday, January 4th, 2016
15:00 - 16:00
Room 011, Kitot Bldg., Faculty of Engineering
Scalable Machine Learning for structured high-dimensional outputs

Abstract:

In recent years, machine learning has emerged as an important and influential discipline in computer science and engineering. Modern applications of machine learning involve reasoning about complex objects like images, videos, and large documents. Treatment of such high-dimensional data requires the development of new tools, since traditional methods in machine learning no longer apply. In this talk I will present two recent works in this direction. The first work introduces a family of novel and efficient methods for inference and learning in structured output spaces. This framework is based on applying principles from convex optimization while exploiting the special structure of these problems to obtain efficient algorithms. The second work studies the success of a certain type of approximate inference methods based on linear programming relaxations. In particular, it has been observed that such relaxations are often tight in real applications, and I will present a theoretical explanation for this interesting phenomenon.

Bio:
Ofer Meshi is a Research Assistant Professor at the Toyota Technological Institute at Chicago. Prior to that he obtained his Ph.D. and M.Sc. in Computer Science from the Hebrew University of Jerusalem. His B.Sc. in Computer Science and Philosophy is from Tel Aviv University. Ofer’s research focuses on machine learning, with an emphasis on efficient optimization methods for inference and learning with high-dimensional structured outputs. During his doctoral studies Ofer was a recipient of Google's European Fellowship in Machine Learning.

~~(The talk will be given in English)

Speaker:   Prof. Yoram Bresler
University of Illinois at Urbana–Champaign

Wednesday, December 30th, 2015
15:00 - 16:00
Room 011, Kitot Bldg., Faculty of Engineering

Learning Sparsifying Transforms for Signal, Image, and Video Processing

Abstract
The sparsity of signals and images in a certain transform domain or dictionary has been exploited in many applications in signal and image processing, including compression, denoising, and notably in compressed sensing, which enables accurate reconstruction from undersampled data. These various applications used sparsifying transforms such as DCT, wavelets, curvelets, and finite differences, all of which had a fixed, analytical data-independent form. Likewise, the acquisition in compressed sensing used mostly random sparse sampling schemes, chosen in a universal way, independent of the data.
Recently, sparse representations that are directly adapted to the data have become popular, especially in applications such as image and video denoising and inpainting. While synthesis dictionary learning has enjoyed great popularity and analysis dictionary learning too has been explored, these methods involve a repeated step of sparse coding, which is NP hard, and heuristics for its approximations are computationally expensive. In this talk we describe our work on an alternative approach: sparsifying transform learning, in which a sparsifying transform is learned from data. The method provides efficient computational algorithms with exact closed-form solutions for the alternating optimization steps, and with theoretical convergence guarantees. The method scales better than dictionary learning with problem size and dimension, a in practice provides orders of magnitude speed improvements and better image quality in image processing applications. Variations on the method include the learning of a union of transforms, and online versions.
We describe applications to image representation, image and video denoising, and inverse problems in imaging, demonstrating improvements in performance and computation over state of the art methods.

Bio: Dr. Yoram Bresler received the B.Sc. (cum laude) and M.Sc. degrees from the Technion, Israel Institute of Technology, in 1974 and 1981 respectively, and the Ph.D degree from Stanford University, in 1986, all in Electrical Engineering. Since 1987, he has been on the faculty at the University of Illinois, Urbana-Champaign, where he is currently Professor, Department of Electrical and Computer Engineering and the Department of Bioengineering, and at the Coordinated Science Laboratory.  Dr. Bresler is also President and Chief Technology Officer of InstaRecon, Inc., which he co-founded in 2003 to commercialize breakthrough technology for tomographic reconstruction developed in his academic research.
Dr. Bresler has served on the editorial board of a number of journals including the IEEE Transactions on Signal Processing, the IEEE Journal on Selected Topics in Signal Processing, Machine Vision and Applications, and the SIAM Journal on Imaging Science, and on various technical committees of the IEEE. He received the IEEE Signal Processing society Senior Best Paper Award in 1988 and in 1989. His honors include the NSF Presidential Young Investigator Award, the Technion (Israel Inst. of Technology) Faculty Fellowship, the Xerox Senior Award for Faculty Research, University of Illinois Scholar, Associate at the Center for Advanced Study of the University, and Faculty Fellow at the National Center for Supercomputing Applications. Dr. Bresler is a fellow of the IEEE and of the AIMBE.
Dr. Bresler’s interests are in multi-dimensional and statistical signal processing and their applications to inverse problems in imaging, and in particular compressed sensing, which he introduced with his students in the mid 90’s under the monikers of “spectrum-blind sampling,” and “image compression on the fly,” as well as computed tomography, magnetic resonance imaging, and learning-based signal processing.