Trust management model that we present is adapted for ubiquitous devices cooperation, rather than for classic client-supplier relationship. We use fuzzy numbers to represent trust, to capture both the trust value and its uncertainty. The model contains the trust representation part, decisionmaking part and a learning part. In our representation, we define the trusted agents as a type-2 fuzzy set. In a decisionmaking part, we use the methods from the fuzzy rule computation and fuzzy control domain to take trusting decision. For trust learning, we use a strictly iterative approach, well adapted to constrained environments. We verify our model in a multi-agent simulation where the agents in the community learn to identify defecting members and progressively refuse to cooperate with them. Our simulation contains significant background noise to validate model robustness.

We study admissible coalitions in goal-directed multiagent systems. We define a qualitative criterion of admissibility in which a coalition has itself all the necessary information to check admissibility. We show also that, under some assumptions on preference relations of the agents, this admissibility criterion can be used to reduce the search space in a game theoretical approach. 1.

General trust management model that we present is adapted for ad-hoc coalition environment, rather than for classic client-supplier relationship. The trust representation used in the model extends the current work by using the fuzzy number approach, readily representing the trust uncertainty without sacrificing the simplicity. The model contains the trust representation part, decision-making part and a learning part.

In this paper we introduce a method to measure the social importance of an agent in a multiagent system, using a directed graph representing dependencies among agents to achieve their goals, so-called dependence graphs pioneered by Castelfranchi, Conte and Sichman. Our measure is derived from van den Brink and Gilles ’ β-measure to rank agents, using a directed graph representing an abstract dominance relation among agents. In particular, we show how to define power structures and dependence networks from the goals and skills of individual agents, and how to adapt the β-measure for such dependence networks based on their topological properties. Moreover, we show that our notion of social importance has a simple and intuitive meaning: it measures the discontent of the other agents in case the agent would leave the society.

The problem of coalition formation when agents are uncertain about the types or capabilities of their potential partners is a critical one. In [3] a Bayesian reinforcement learning framework is developed for this problem when coalitions are formed (and tasks undertaken) repeatedly: not only does the model allow agents to refine their beliefs about the types of others, but uses value of information to define optimal exploration policies. However, computational approximations in that work are purely myopic. We present novel, non-myopic learning algorithms to approximate the optimal Bayesian solution, providing tractable means to ensure good sequential performance. We evaluate our algorithms in a variety of settings, and show that one, in particular, exhibits consistently good sequential performance. Further, it enables the Bayesian agents to transfer acquired knowledge among different dynamic tasks.

Coalition formation is an important form of interaction in multiagent systems. It enables the agents to satisfy tasks that they would otherwise be unable to perform, or would perform with a lower efficiency. The focus of our work is on real-world application domains where we have systems inhabited by rational, self-interested agents. We also assume an environment without any trusted central manager to resolve issues concerning multiple agents. For such environments, we have to determine both an optimal (utilitymaximizing) coalition configuration and a stable payoff configuration, concurrently and in a distributed fashion. Solving each of these problems is known to be computationally expensive, and having to consider them together exacerbates the problem further. In this paper, we present our Progressive, Anytime, Convergent, and Time-efficient (PACT) algorithm for coalition formation to address the above concerns. We assess the stability of the resulting coalition by using a new stability concept, the relaxed core, which is a slight variation on the core. We show experimentally that our algorithm performs admirably in comparison to an optimal solution, it typically produces solutions that are relaxedcore-stable, and it scales well. 1.

Sequential decision making under uncertainty is always a challenge for autonomous agents populating a multiagent environment, since their behaviour is inevitably influenced by the be-haviour of others. Further, agents have to constantly struggle to find the right balance between exploiting current information regarding the environment and the rest of its inhabitants, and ex-ploring so that they acquire additional information. Moreover, they need to profitably trade off short-term rewards with anticipated long-term ones, while learning through interaction about the environment and others—employing techniques from reinforcement learning (RL), a fun-damental area of study within artificial intelligence (AI). Coalition formation is a problem of great interest within game theory and AI, allowing autonomous individually rational agents to form stable or transient teams (or coalitions) to tackle an underlying task. Agents participating in realistic scenarios of repeated coalition formation under uncertainty face the issues identified above, and need to bargain to succesfully negotiate the terms of their participation in coalitions—often having to compromise individual with team welfare effectively. In this thesis, we provide theoretical and algorithmic tools to accommodate sequential de-

Abstract. In many applications of the task allocation problem such as peer-topeer and grid computing, and virtual organizations, the (social or business) relations between the participating agents play an important role, and thus they should be taken into account. Furthermore, in such applications, agents providing the resources usually act self-interested. This paper therefore studies the problem of finding truthful mechanisms for these kinds of social task allocation problems. In this paper we give on the one hand an optimal mechanism and model the problem as an integer linear program (ILP), and on the other hand a polynomialtime approximation by splitting the problem into smaller sub-problems, each of which is solved optimally. We show that both mechanisms are truthful. The optimal mechanism may take exponential time for some instances, and in theory, the quality of the approximation is not guaranteed. However, we show experimentally that for problem instances where the social network has the smallworld property, the quality of the results for the approximation is quite good, due to the fact that the division into subproblems uses the locality of tasks in the social network. 1