Title and Abstract:
Compressed Sensing and Search-based Sparse Recovery Methods
Compressed sensing (CS) has been an emerging topic in the signal processing community in the last decade. In contrast to the conventional first-capture-then-compress methods including the well known standards such as JPEG and MPEG, CS aims at combining the signal acquisition and compression steps. Exploiting sparsity or compressibility of signals, CS enables acquisition at rates lower than the Nyquist rate. This is accomplished by obtaining a lower dimensional set of observations instead of the higher dimensional original signal.
CS capturing process necessitates recovering the original signal of interest from the observations, which is an ill-posed problem due to the reduction of dimensions. Casting this as a sparsity-promoting optimization, it is possible to define a recovery problem with a unique solution under some constraints such as the restricted isometry property. Despite the uniqueness of the solution under such conditions, the resultant sparsity-promoting optimization problem requires an intractable search, which has motivated the use of approximate methods. In the last decade, a vast number of methods has been developed in the CS community to solve the sparse recovery problem. Though they may also be grouped in different ways, some of these recovery methods, including orthogonal matching pursuit, forward-backward matching pursuit, A* orthogonal matching pursuit, etc. can be categorized as search-based techniques.
In this talk, first, the motivation behind CS will be discussed. Common sparse recovery methods and theoretical guarantees for exact recovery will be visited. Some recent extensions and applications of CS will be mentioned. In the second part of this talk, search-based recovery techniques will be introduced. Their recovery performances will be outlined by a number of recovery examples comparing them to the conventional recovery techniques. Recovery performance will also be demonstrated on images as well.
Short biography of the speaker:
Dr. Nazım Burak Karahanoğlu is a senior researcher at the Advanced Technologies Research Institute of the Scientific and Technological Research Council of Turkey (TUBITAK) in Kocaeli, Turkey.
He received his B.S. degree in Electrical and Electronics Engineering from Middle East Technical University, Ankara in 2003. He got his M.S. degree in Computational Engineering in 2006 from the Friedrich-Alexander University of Erlangen-Nuremberg, Germany, where he conducted research in speech signal processing, speech recognition and beamforming. He worked for a year at the Friedrich-Alexander University in an EU funded project on Distant Talking Interfaces for Digital TV before joining TUBITAK in 2008. He completed his Ph.D. studies in Electronics Engineering at Sabancı University under the supervision of Asst. Prof. Dr. Hakan Erdoğan between 2008 and 2013. His research focus was compressed sensing and search-based sparse recovery algorithms. Since 2008, he has been with TUBITAK, where he is the team leader of a group working on sonar signal processing and simulation.
His research interests include compressed sensing and sonar signal processing.