22nd International Symposium on Computer and Information Sciences	November 7-9, 2007
Middle East Technical University, Cultural and Convention Center, Ankara, Turkey

KEY SPEAKERS

Erol Gelenbe, Imperial College, UK Networked Auctions We present a model of sellers and bidders that interact through a networked auction system. The analysis of the system provides closed form expressions for the number of bidders present at each of the auction sites, and for the price that is attained at each site. It also provides a way to compute the optimum selling policy that each seller must apply in order to obtain the highest income per unit time, and for bidders to select routing policies which optimise their own self interest. The system we describe is being currently implemented on a self-aware network test-bed at Imperial College. Nicholas R. Jennings, University of Southampton, UK Decentralised Control of Complex Systems Many modern computing systems have to operate in environments that are highly interconnected, highly unpredictable, are in a constant state of flux, in which there is no centralized point of control, and in which the constituent components are owned by a variety of stakeholders that each have their own aims and objectives. Relevant exemplars include the Web, Grid Computing, Peer-to-Peer systems, Pervasive Computing and many eCommerce applications. Now, I believe that all of these systems operate under the same conceptual model: (i) entities offer a variety of services in some form of institutional setting; (ii) other entities connect to these services (covering issues such as service discovery, service composition and service procurement); and (iii) entities enact services in a flexible and context sensitive manner. Moreover, I believe agent-based computing is an appropriate model computational for such systems (Jennings, 2000; Jennings 2001). In particular, autonomous agents are a natural way of viewing flexible service providers and consumers and the interactions between these autonomous components are naturally modeled as some form of economic trading process that, if successful, results in a service contract (or service level agreement) between the agents involved. In this talk, the focus will be on the design of the agents and their interactions. Specifically, I will consider the design and implementation of various computational service economies for a number of real-world applications including: virtual organizations (Norman et al., 2004), sensor networks (Padhy et al., 2006; Rogers et al., 2005; Rogers et al., 2006) and personalized recommendations (Wei et al., 2005; Payne et al., 2006). In so doing, I will touch upon some of the techniques and advances we have made in the areas of game theory (Gerding et al., 2006), auctions (Dash et al., 2007, David et al., 2005), coalition formation (Dang and Jennings, 2006; Dang et al., 2006; Rahwan et al., 2005; Rahwan and Jennings, 2005), automated negotiation (Fatima et al., 2004; Fatima et al., 2006) and computational mechanism design (Dash et al., 2003). Ozgur Ulusoy, Bilkent University, Turkey Research Issues in Peer-to-Peer Data Management Data management in Peer-to-Peer (P2P) systems is a complicated and challenging issue due to the scale of the network and highly transient population of peers. In this paper, we identify important research problems in P2P data management, and describe briefly some methods that have appeared in the literature addressing those problems. We also discuss some open research issues and directions regarding data management in P2P systems. Keywords: Peer-to-peer systems, data management, overlay network, indexing, data integration, query processing, data replication, clustering, free riding, incentive mechanism.

INVITED TALKS

NETWORKS AND SYSTEMS TRACK

Nail Akar, Bilkent University,Turkey Teletraffic Aspects of Optical Burst Switching Networks Optical burst switching is a packet switching-based paradigm that has recently been introduced to cope with bursty traffic and for more efficient utilization of the fiber capacity in wavelength division multiplexing-based optical networks. It is generally believed that optical burst switching is one of the most promising candidates for the transport technology of the next generation Internet. In electronically switched networks, the use of electronic buffering makes it possible to get the most use of the fiber transmission capacity. For such networks, queueing theory deals with buffer dimensioning, sizing, and engineering problems using teletraffic models. However, general buffering is not possible with current technologies for optical burst switching networks and only a portion of the fiber transmission capacity can be used to carry data. The reason behind this throughput reduction is the so-called contention which arises as a result of two or more optical bursts contending for the same output wavelength. Contention can be resolved in wavelength domain by wavelength converters, in time domain by fiber delay lines or in space domain by deflection routing. In this talk, an overview of optical burst switching and contention resolution methods will be given and analytical teletraffic models will be presented to quantify the throughput achievable in optical burst switching networks. Cem Ersoy, Bogaziçi University, Turkey Wireless Sensor Networks Wireless Sensor Networks (WSNs) are special ad-hoc networks, which are being proposed for many civil and military applications. Especially, in the aftermath of natural disasters, where the existing infrastructure is totally unusable, and civil and military sensing applications that requires no infrastructure belong to those group of applications. Via employing those low cost-producible sensors, it is possible to produce simple solutions for many serious problems. For example, by forming a network of temperature sensors on vast forest lands, it is possible to locate fires during their initial phases and take necessary precautions and take the fire under control as early as possible. Similarly, by using similar methods it is possible to prevent floods. For regions that require irrigation, sensors that measure the changes in the salinity ratio of the clay will make it possible to take necessary measurements for local problems as early as possible. Keeping detailed track of the changes in the local climate for dam-regions will help the local authorities to decide on the most appropriate agricultural investments for that region. It is possible to group the related scientific research in two branches i) design and development of wireless sensors ii) design of the wireless sensor network and development of the communication protocols for the wireless sensor networks. The output from the former section of research branch is wireless sensors which are of the size of a coin, and they are trying to develop even smaller sensors. But smaller sizes create battery capacity problems and consequently communication limits, which boosts the importance of the second branch of research. So, the main aim of the second branch is to design the network and develop the communication protocols that will enable the wireless sensor network have the longest operational lifetime under limited battery capacities. This talk summarizes the research activities on WSN in NETLAB of Bogazici University. Described problems here are related to the design of event triggered wireless sensor networks for detecting events like forest fires, floods, avalanche and anticipating earthquakes, and nuclear, chemical and biological attacks and intrusion detection and time-triggered sensor networks for data gathering of the condition of the soil for erosion, humidity, salinity, condition of the seeds on big agricultural vast lands. Development of energy efficient communication protocols, evaluation of the performance of the designed wireless sensor networks via OPNET simulations and experiments on small size testbeds are presented. Semih Bilgen, Middle East Technical University, Turkey Recent Developments in Networking Technology In this talk, recent research work in computer networking will be briefly reviewed. First recent developments in wide area networking research will be overviewed. The ten years of optical networking research will be outlined, with emphasis on recent work on optical burst switching, IP over DWDM. In this part recent topics such as IP multimedia systems amd the general fusion of software engineering and networking technologies, as exemplified in such applications as peer to peer video streaming and objects methodologies applied in networking development will also be addressed. Then, access networking topics will be considered. Recent work on EPON and developments in the direction of 100GbpsEthernet will be summarized, after a brief history of the work on "the last/first mile". Wireless networking will be the next subject to be overviewed. Fourth generation networking topics will be summarized and recent work on mobile and ubiquitous networking will be outlined. Sensor and ad-hoc networking research will also be considered as this has constituted a popular branch of networking research, especially in the last decade. Finally, an attempt will be made to overview the recent research undertaken in Turkey in these areas. Some research projects known to be under way in major Turkish research institutions will be briefly mentioned. The talk will end with a humble look towards the future of networking research in the world as well as in Turkey.

COMPUTER VISION, GRAPHICS AND INTELLIGENCE TRACK

Cem Say Qualitative Reasoning Meets Computability Theory Qualitative reasoning (QR) researchers develop AI programs which use "weak" representations (like intervals rather than real numbers for quantities, and general shape descriptions rather than exact formulae for functional relationships) in their vocabularies to perform various reasoning tasks about incompletely specified dynamical systems. Qualitative simulation is an important QR technique with numerous applications. The task of a qualitative simulator is to compute all the qualitatively distinct possible behaviors of a system whose model and initial state description are given using very little precise information. In recent years, we used the tools of computability theory to make some interesting discoveries about the capabilities of qualitative simulators: We now know that it is impossible to build a qualitative simulator which will predict all and only the realizable behaviors of all possible input systems, that this problem does not go away even if one restricts the modeling language in a significant way, and that every sufficiently good qualitative simulator Q has a "blind spot," i.e. an input model containing an inconsistency which can be detected easily by many other simulators, but which will fool Q into simulating it to produce a spurious prediction. We will give a brief overview of our results and future directions. Veysi Isler, Middle East Technical University, Turkey Research Challanges in Interactive Computer Graphics Increase in the computing power enables more advanced graphics algorithms to get utilized in interactive applications. The related hardware technology rising to this increase comes out mostly in the form of paralel and/or distributed architectures. For instance, the graphics boards with gpu (graphical processing units) avaliable for most of the desktop computers today contain a kind of paralel streaming processors. On the other hand, there are now more distributed interactive applications such as massively multiplayer 3D games and 3D collaborative interactive environments which are based on paralel and distrbuted architectures. Change in the underlying computer hardware technology towards parallel and distributed architectures for graphics applications necessitates reconsidering and redesigning of algorithms and methods with the employed data structures, accordingly. In this seminar, we present a survey of research studies and challenges in interactive computer graphics which are based on parallel and distributed computing. Lale Akarun, Bogazici University, Turkey 3D Face Recognition and Advances and Challanges 3D face recognition has solved many of the problems of 2D face recognition: 3D is less affected from illumination and pose changes, which plague 2D face recognizers. 3D face recognition systems have matured to a degree where recognition performance is at least as good as 2D face recognizers, and many 3D sensors have become widely available, if not commonly affordable. In this talk, an overview of the techniques used in 3D face recognition will be overviewed, some example algorithms will be surveyed, and reamining challanges will be presented.

DATA MANAGEMENT TRACK

Fazli Can, Bilkent University, Turkey New Event Detection and Tracking: Test Collection Construction, Algorithms, and Evaluation New event detection and tracking (NEDT) systems aim at identifying the first stories corresponding to new events and clustering news articles tracking these first stories. NEDT systems use news articles provided by a variety of Web resources. Commercial Web application examples of such systems include Google News and NewsIsFree. Research and development of these systems need a lab environment that involves test collections. These collections include a set of chronologically ordered news articles. In these articles, human annotators identify the first stories and the corresponding tracking news. Using test collections, different NEDT algorithms and their associated parameters can be tuned to attain a desirable system performance. In this paper, we illustrate the construction of a NEDT test collection, and the development and evaluation of various NEDT algorithms. Index Terms: information retrieval, new event detection and tracking, test collection construction, Web news resources. Acknowledgment This work is carried out with Seyit Kocberber, Ozgur Baglioglu, Suleyman Kardas, H. Cagdas Ocalan, and Erkan Uyar. All are members of the Bilkent Information Retrieval Group. Attila Gursoy, Koc University, Turkey Applications of Data Mining and Integration to Study Protein Interactions. In the last decade, an interesting convergence of Biology and Computer Science has taken place: whereas Biology has become a more information-driven science, Computer science has moved towards natural sciences, as it studies more and more fundamental information processes found in nature. The need to manage, mine, integrate, and interpret vast amount of biological data has become even more significant today as more genome sequencing projects result in exponential growth of diverse, fragmented, and distributed biological data. One of the primary objectives of the post-genomic era is the elucidation of the interactions in cellular systems. Such knowledge aids researchers in identifying nodes in biological pathways that cause disorders, for example cancer, and designing drugs that exert therapeutic action on these nodes. This talk presents applications of data mining and data integration techniques for the prediction of gene and protein interactions. Various data sources such as protein structures and microarray expression profiles are integrated for a better understanding of biological pathways.