Bo�azi�i �niversitesi Bilgisayar M�hendisli�i B�l�m�'nden Haberlerhttp://www.cmpe.boun.edu.tr/index_tr.phpAsst. Prof. Dr. Abu Asaduzzaman

Modern academic research activities and information technology (IT) related industry challenges require low-power high performance computing, which can be achieved through simultaneous multithreading (SMT) and graphics processing unit (GPU)-assisted parallel computing. The objective of this presentation is to introduce SMT-capable multicore central processing unit (CPU) and some state-of-the-art GPU-accelerated parallel programming techniques, analyze some highly computation intensive applications, and share some experimental results explaining the impact of SMT/GPU on performance to power ratio of multicore/manycore systems. Based on the Laplace�s Equation implementation for 2D electric charge distribution in our Computer Architecture and Parallel Programming Laboratory (CAPPLab), SMT/GPU-based parallel computing may achieve up to 30x speed up and save up to 96% energy consumption for a 4096x4096 thin surface. Finally, this presentation presents WSU CAPPLab � its researchers, resources, and activities. CAPPLab stands ready to collaborate with other institutions in any possible research endeavor

Tarih: 02.06.14 12:00
Yer: AVS Seminer Odas�]]>Mon, 02 Jun 2014http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=277 Theo Gevers / University of Amsterdam

In this talk, I will discuss various recent developments and state-of-the-art methods in color-based object recognition. The focus is on discussing invariance properties and the distinctiveness of interest point detectors and color descriptors for object recognition and localization. Then, these methods are used to provide human activity recognition and 3D reconstruction/printing.

Short Bio: Theo Gevers received his Ph.D. degree in Computer Science from the University of Amsterdam in 1996. He is a Professor of computer science with the University of Amsterdam, Amsterdam, The Netherlands, where he is also a Teaching Director of the M.Sc. degree in artificial intelligence. He has published over 100 papers on color image processing, image retrieval, and computer vision. His main research interests include the fundamentals of content-based image retrieval, color image processing, and computer vision, specifically in the theoretical foundation of geometric and photometric invariants. Dr. Gevers currently holds a VICI award (for excellent researchers) from the Dutch Organisation for Scientific Research. He served as program committee member, invited speaker, and lecturer of postdoctoral courses at various major conferences, such as the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), International Conference on Pattern Recognition (ICPR), The international society for optics and photonics (SPIE), and European Conference on Colour in Graphics, Imaging, and Vision (CGIV). He is a co-chair of the European Conference of Computer Vision (ECCV) 2016.

Tarih: 30.05.14 11:00
Yer: AV� - ETA A 16]]>Fri, 30 May 2014http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=276 Prof. Oguzhan Ocan, Design Lab, Koc University



Tarih: 20.05.14 12:00
Yer: AVS Seminar Room (No:16)]]>Tue, 20 May 2014http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=275 Kuban Alt�nel, Bogazici University, Industrial Eng.

Sensors are autonomous tiny devices with limited energy and capability for sensing, data processing and communication; but they can collectively behave to provide an effective wireless sensor network that monitors a region, and transmits information to gateway nodes called sinks. In most of the applications, the network must operate for long periods of time, which makes the management of the available energy resources of sensors an important issue. In this work, we first present a mixed-integer linear programming model that maximizes the network lifetime by optimally determining sensor and sink locations, sensor-to-sink data routes and activity schedules of the deployed sensors over a finite planning horizon subject to coverage, flow conservation, energy consumption and budget constraints, and introduce valid inequalities to strengthen the formulation. Then, we propose a reformulation and a method that efficiently solves its Lagrangean dual using column generation and subgradient optimization, which we embed within a branching scheme to compute an optimal solution. Based on the computational results we can say that the proposed valid inequalities increase the strength of the formulation. Also the developed branch-and-bound algorithm gives either the optimal solution or a near optimal solution with a very small optimality gap within the imposed time limit for large instances.
Co-authored with Yavuz Bo�a� T�rko�ullar�, N. Aras and Cem Ersoy


Tarih: 01.04.14 12:00
Yer: AVS, (CMPE Seminar room ETA 16)]]>Tue, 01 Apr 2014http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=274 Mohan Kankanhalli, National University of Singapore

ABSTRACT:
There has been an increasing research interest in large-scale distributed camera networks. In many of these systems, the multiple sensors operate separately in isolation and only the central processing unit fuses the data obtained from the various sensors to accomplish its task. However, if we can coordinate and control these sensors, the system resource utilization can significantly improve.

We will first present some current work focusing on decision-theoretic approaches for control and coordination for multi-camera systems, where there are uncertainties due to sensor noise, vision algorithms and non-determinism of the environment. Decision-theoretic approaches use utility functions to maximize the number of surveillance targets being tracked or to provide fairness in the observation times.

We will then describe some of our very recent work in social interaction detection using multi-camera systems. We detect the occurrence and location of social interactions via extended F-formation detection, which is a concept from sociology that is used to define social interactions. We use information from visual cameras and Kinect cameras to build a temporal extension of the F-formation system that can help detect social interactions, which is very useful in many applications.

BIOGRAPHY:
Mohan Kankanhalli is a Professor at the Department of Computer Science of the National University of Singapore. He is also the Vice Provost for Graduate Education at NUS. Before becoming the Vice Provost in 2014, he was the Associate Provost (Graduate Education) during 2011-2013. Earlier, he was the Vice-Dean for Academic Affairs & Graduate Studies at the NUS School of Computing during 2008-2010 and Vice-Dean for Research during 2001-2007. Mohan obtained his BTech from IIT Kharagpur and MS & PhD from the Rensselaer Polytechnic Institute.

His current research interests are in Multimedia Systems (content processing, retrieval) and Multimedia Security (surveillance and privacy). He has been awarded a S$10M grant by Singapore's National Research Foundation to set up the Centre for "Sensor-enhanced Social Media" (sesame.comp.nus.edu.sg).

Mohan is actively involved in organizing of many major conferences in the area of Multimedia. He is the Technical Program Co-Chair for ICMR 2014. He was the Director of Conferences for ACM SIG Multimedia during 2009-2013. He is on the editorial boards of several journals including the ACM Transactions on Multimedia Computing, Communications, and Applications (TOMCCAP), Springer Multimedia Systems Journal, Pattern Recognition Journal and Multimedia Tools & Applications Journal. He received the TOMCCAP Associate Editor of the Year Award in 2013. He is a Fellow of IEEE.


Tarih: 21.02.14 16:00
Yer: AV� - ETA 16]]>Fri, 21 Feb 2014http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=273 Nicholas Wisniewski Department of Medicine University of California, Los Angeles

Abstract
How different are two statistical models inferred from a finite number of measurements? How do we assess if a difference is distinguishable by measurement? These very general questions are quite common in statistical and engineering applications such as signal processing and computer vision. Although there are some techniques appropriate to particular situations, current statistical methodology does not generalize across arbitrary families of probability distributions. New statistical measures that are generalizable have emerged from the concept of information distance in probability spaces, as formulated in the language of differential geometry. These distances have received some attention in the scientific literature, and there have been theoretical remarks connecting information geometry and hypothesis testing, but there exists little framework connecting the geometric theory to statistical practice, apparently due to mathematical difficulties in finding explicit solutions. By making use of numerical integration methods, we develop the necessary connections between theory and methodology that allow the information geometric distance to be used as an effect size statistic. In particular, we define the confidence interval for the information geometric distance, introduce a general algorithm to compute it, and extend its use into a framework of null hypothesis significance testing and power analysis for an arbitrary family of probability distributions.

Short Bio
Nicholas Wisniewski received his B.S. in physics from Caltech in 2001, where he worked on algorithmic development in the search for the Higgs boson at CERN, Geneva, Switzerland. As a graduate student, he began studying behavior in complex systems, and became interested in applications to biology and medicine. He received his Ph.D. in physics from the University of California, Los Angeles in 2010, using information geometry to model inferences made in diffusion tensor magnetic resonance imaging (DT-MRI), and developing algorithms for reconstructing tensor fields. As research faculty at UCLA, he is currently involved in developing machine learning techniques for the analysis of biological networks, as well as developing models of morphogenesis in clonal growth patterns from stem cells. He also teaches his own class in biostatistics to graduate students in bioinformatics and life sciences.


Tarih: 06.01.14 11:00
Yer: AVS, (CMPE Seminar room ETA 16)]]>Mon, 06 Jan 2014http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=272 �smail Ar�, Bogazici University

Transcription of polyphonic piano music is an important computer music problem and many sophisticated methods have been proposed for its solution. However, most techniques cannot fully utilize all the available training data efficiently and do not scale well beyond a certain size. We develop an exemplar-based approach that can easily handle very large training corpora. We maintain transcription performance by only retaining 1% of the training data. The method is competitive with the state-of-the-art techniques in the literature. Besides, it is very efficient and can work in real time.

Tarih: 22.11.13 10:30
Yer: AVS Seminar Room]]>Fri, 22 Nov 2013http://www.cmpe.boun.edu.tr/announcements/showWeeklyProgram_tr.php?eventid=271






Bilim S�nav�: 9 May�s, 2014, Cuma, 9:00-15:00, CMPE binas�]]>Tue, 04 Mar 2014http://www.cmpe.boun.edu.tr/announcements/index_tr.php?id=1393935033


Tarih: 10 Subat 2014 Pazartesi 13:00



Yer: Bilgisayar M�hendisli�i B�l�m� ETA A5 S�n�f�]]>Thu, 23 Jan 2014http://www.cmpe.boun.edu.tr/announcements/index_tr.php?id=1390479048 Tue, 20 Aug 2013http://www.ceng.metu.edu.tr/~acm/SummerCamp2013/