CS: Image reconstruction by deterministic CS, Compressed Signal Reconstruction using the Correntropy Induced Metric, a PhD job and a talk

Found on the interwebs:

Image reconstruction by deterministic compressive sensing by Kangyu Ni, Somantika Datta, Svetlana Roudenko, Douglas Cochran. The abstract reads:

A recently proposed approach for compressive sensing with deterministic measurement matrices is applied to images that possess varying degrees of sparsity in their wavelet representations. The use of these deterministic measurement matrices is found to be approximately as effective as the use of Gaussian random matrices in terms of image reconstruction fidelity. The ``fast reconstruction'' algorithm enabled by this deterministic sampling scheme produces accurate results, but its speed is hampered when the degree of sparsity is not sufficiently high.

I mentioned Compressed Signal Reconstruction using the Correntropy Induced Metric by Sohan Seth, Jose C. Principe earlier. Now Sohan Seth also released the attendant poster and the code implemented in the article. I will add it to the reconstruction section of the big picture.

Paul Honeine, an assistant professor at University of Troyes in France has an opening for a Ph.D. studentship in  Compressive sensing for pattern recognition and decision making in wireless sensor networks. The summary of his ad reads:

Over the past few years, wireless sensor networks received tremendous attention for monitoring physical phenomena, such as the temperature field in a given region, and for tracking a mobile unit. With relatively inexpensive wireless devices, each device has a limited amount of memory, reduced processing capabilities, limited power resources, and low communication capacities. In order to carry out good coverage of the region under scrutiny, sensors must be deployed densely for collaborative processing data, resulting in highly redundant information. Exploiting the redundancy in signal representations has very recently received a great deal of attention in the signal processing community, with the compressive sensing approach, by optimizing both the acquisition and the compression steps. Clearly, wireless sensor networks still has not taken advantage of these new methods, although many efforts have been focused to develop signal representations.

The PhD candidate will study the problem of learning sparse representations for wireless sensor networks, with compressive sensing techniques. As these methods employ random projections of the data as a compression scheme, this can be exploited to control the energy resource within each sensor with smart sleeping policies. He will define such policies by studying criteria such as the

sensing capacity for a large class of applications involving monitoring a physical phenomenon as a functional estimation problem, and detecting and tracking as a decision rule. The PhD candidate will take advantage of recent developments in kernel-based machine learning and sparse representations within the team he will join.

Advisor : Paul HONEINE, assistant prof. (paul.honeine@utt.fr)

Lab. : LM2S, University of Technology of Troyes (lm2s.utt.fr)

I'll add this announcement to the CS jobs section.

The IPSN 2009, The 8th ACM/IEEE International Conference on Information Processing in Sensor Networks will take place on April 13-16, 2009 in San Francisco. The meeting will host a Workshop on Compressive Sensing Theory and Practice (CSTP 09).

The rest of the post is in french. Let me just say I find it odd that Compressed Sensing is not listed in any of the talks given at the all day meeting organized by the SEE at the French Academy of Science. 

Au Séminaire Cristolien d'Analyse Multifractale, Centre de Mathématiques, Faculté de Sciences et Technologie, Université Paris XII - Val de Marne, Patrick Flandrin (ENS Lyon) donnera une presentation initulee Une approche "compressed sensing" pour la localisation temps-fréquence. Summary in French, Jeudi 29 Janvier 2009 : Salle P2-132 (Bâtiment P2, 1er étage), 13h30-14h30. Resume:

Une approche ”compressed sensing” pour la localisationtemps-frequence. La representation temps-frequence ”ideale” d’un signal AM-FM multi-composantes est par nature parcimonieuse dans la mesure ou elle se resume essentiellement a un ensemble de trajectoires 1D (ponderees) n’occupant qu’une fraction reduite du plan. Les progres recents apportes par les techniques de ”compressed sensing” aux problemes de reconstruction sous contrainte de parcimonie suggerent de revisiter sous un angle nouveau la question de la localisation temps-frequence dans le contexte specifique des distributions d’energie. En se basant sur des resultats classiques et d´eja anciens selon lesquels l’information pertinente relative a la localisation des composantes individuelles d’un signal se concentre au voisinage de l’origine du plan de sa fonction d’ambiguıte, on montrera qu’une distribution temps-frequence tres localisee peut etre obtenue comme la solution a norme l_1 minimum d’un probleme d’optimisation contraint par un nombre tres restreint de valeurs dans le domaine de sa transformee de Fourier. On illustrera les possibilites et limitations de cette nouvelle approche dans diverses configurations, en les comparant aux methodes classiques. On montrera ainsi que des performances accrues sont possibles, quoiqu’a un cout algorithmique sensiblement augmente.

I'll add these to the CS Calendar soon.

Image Credit: NASA/JPL/Space Science Institute, Saturn as seen on January 5th, 2009.