Technical Program

Constrained Coalitional Games and Networks of Autonomous Agents
Prof. John S. Baras, University of Meryland

In this talk we develop a unifying analytical and optimization framework for the design, operation and performance evaluation of networks of autonomous agents. The fundamental view is that agents in such a network are dynamic entities that collaborate because via collaboration they can accomplish objectives and goals much better than working alone, or even accomplish objectives that they cannot achieve alone at all. Yet the benefits derived from such collaboration require some costs (or expenditures), for example due to communications. Or in equivalent terms, the collaboration is subject to constraints (could also be dynamic). Understanding and quantifying this trade-off between the benefits vs the costs of collaboration, leads to new methods that can be used to analyze, design and control/operate networks of agents. Multiple metrics for benefits and costs can be considered within this framework. We present results on the significant effects of network topology on performance. In all cases distributed and partially asynchronous algorithms and solutions using local information are emphasized.

Eigenvector Based Reputation Measures
Danil Nemirovsky, INRIA Sophia Antipolis

Reputation systems are indispensable for the operation of Internet mediated services, electronic markets, document ranking systems, P2P networks and Ad Hoc networks. In the tutorial we recall notions of trust and reputation and, particularly, eigenvector based reputation measures which are used in a number of applications like Web and P2P networks. Also we consider available distributed approaches to the eigenvector based reputation measures. We classify these approaches into two categories. The first category is based on asynchronous methods. The second category is based on the aggregation/disaggregation methods.

Sharing Resources in Grids
Prof. Costas Courcoubetis, Athens University of Economics and Business

We consider a problem of dimensioning and paying for a pool of resource that is to be shared by n users. We have in mind, for example, a computing resource pool that is to be shared amongst users who participate in Grid computing. At the start, each potential user reveals some private information about the utility he can obtain by being granted resource from the pool (the type of the user). On the basis of these revelations three things are determined: how large the pool is to be, how much each user is to pay towards it, and how the resource of the pool is to be shared when several users call upon it on the same day, with the additional modelling assumption that users announce on each given day some further information about the utility that they have for resource that particular day. Knowing rules for how all these parameters of the system are to be determined (as a function of all the user revelations), each user will choose what he reveals in a self-interested way. Our job is to design the rules so the system operates with the maximum total net benefit that it can, given that all users act self-interestedly. In this talk we explore interesting questions about the optimal design. For instance, is it true that on a given day the allocation of the resource amongst the users who contend for it is efficient, i.e., such that the sum of these users' net benefits is maximized on that day? The answer to this question has an important implication as to the questions of the usefulness of using an internal market to allocate resources in a Grid computing system.