George Vouros1, Nikos Vlassis2 (Area Editors)
1University of the Aegean, Dept. of Information and Communication Systems Engineering,


2Technical University of Crete, Dept. of Production Engineering and Management,













Results in agents and multi-agent systems (AMAS) research are likely to lead many developments in areas of information and communication technology, penetrating many areas of industry and services. Robotics, Cognitive Systems, Peer to Peer Systems, Grid Computing, Semantic Web, Manufacturing, Mobile and Ubiquitous Computing are just some of the important areas, where developments in AMAS have tremendous impact.

Agents have brought a new metaphor in computing that, although new, is maturing fast, allowing the development and study of complex systems that would be extremely difficult to engineer, deploy and study in a pre-agents era. However, there are many things that need to be studied and understood as far as agent-based systems are concerned: Something that is evidenced by the number of active researchers world-wide, well known conferences, workshops, interest of funding bodies and thus of industry.

In the next paragraphs we shall briefly provide basic terminology, the context and the goals of research in agent-based systems: Agents are autonomous entities that, being in constant connection with their environment, are capable of flexible, reactive, goal-directed and interactive behavior, even in difficult environments. The environment includes agents’ physical environment (i.e. the environment where they are deployed and where they bring changes by performing their actions), as well as other agents existing in it. This environment may be quite “simple” or “difficult”, depending on agents’ awareness-capacity (as this is determined/imposed by agents’ perception abilities and by the environment itself), environment’s rate of change, agents’ knowledge of changes’ causalities, and by other agents acting on it. This, in conjunction to the complexity of the tasks, challenges individual, as well as collective agents’ abilities: Their abilities to reason, plan and act, to continuously learn, as well as to gather themselves into coalitions/groups/organizations – functioning wholes-, to collaborate so as to achieve their goals by sharing information. Collaboration among others entails reconciling differences, reaching agreements, negotiating and acting in a coordinated fashion. Furthermore, to engineer multi-agent systems that act effectively in difficult environments and perform complex tasks, it is important for agents to preserve constant and consistent awareness of the situations arising, adapting (either as individuals or as groups/organizations/wholes) to the changes arising, and “tuning” their behavior to the changes and needs that (continuously) arise.

As it is stated in [LMS+05], agent technologies can be viewed from three perspectives: As a way for engineering applications around autonomous communicative components, as media for representing complex and dynamic real-world environments, and as a source of new technologies related to the abilities of agents. These constitute the goals of research in agent-based systems. Research efforts are being conducted in a context that necessitates technological solutions for the development of distributed systems. These technological solutions (mobile computing, sensor networks, P2P systems, semantic web, SOAs, e-commerce) highly interact with developments in agent-based systems.

Viewing agent technologies as sources of technological developments, this report aims to present efforts in Greek sites, in Laboratories, teams and by individuals, to advance agent technologies. The text is structured along thematic areas, as detailed below.



Formal mathematical foundations for multi-agent systems received much attention from researchers in the field over the last decade. A special class involves norm-governed (normative) multi-agent systems. A characteristic feature of these systems is that actuality, what is the case, does not always coincide with ideality, what ought to be the case. Members of such systems may fail to, or even choose not to, conform to the system specifications. Consequently, the behaviour of the member-agents needs to be regulated by a set of laws expressing their permissions, obligations, rights, duties, and other, possibly more complex normative relations that may exist between them.

In this direction, researchers from the Institute of Informatics & Telecommunications, at the National Centre for Scientific Research (NCSR) "Demokritos", in Athens, Greece, have developed an executable specification of norm-governed multi-agent systems [ASP07, ASP08, APS02]. The specification explicitly represents: (i) the constitutive laws of a system, (ii) the normative environment, and (iii) the physical environment within which the agent interactions take place. Constitutive laws define the meaning of the agents' actions while the normative environment expresses the agents' permissions, obligations, rights, and so on. They formalise the executable specification with the use of action languages from the field of Artificial Intelligence; they have used the action language C+, developed by the Action Group of the University of Texas, and the Event Calculus, developed at Imperial College London. They have demonstrated their approach by specifying and executing protocols for argumentation, e-commerce, coordination, resource sharing, and voting.

While statically specified Norm-Governed AMAS are appropriate for a wide range of systems, there are several cases where for environmental, social or other reasons, dynamic specifications through their runtime modification are desired, or, indeed, essential. Examples of such systems include long-running virtual organizations operating under evolving legal and/or social frameworks, the adaptation of a voting protocol at runtime with the aim of defeating its manipulation through strategic voting and rerouting data through the adjustment of the routing protocol itself in a wireless sensor network deployed in harsh and volatile environmental conditions.

In addressing this need, researchers from the above group have been working towards an extended framework for dynamic specifications. The framework explicitly supports runtime modification of the specifications, while retaining many of the characteristics of the static counterpart. Specifications are modified by the participating agents through the initiation of meta-protocols whose effects include the addition, removal or replacement of rule-sets. In addition to the framework, they have been working towards a mathematical model and related software aiming to assist both in the design and evaluation of such systems by presenting a quantifiable account of the evaluation criteria [KP07].

In a related line of research, researchers from the Intelligence, Modelling and Computation (IMC) group of CITY College, Thessaloniki, have focused on formal modeling of agents, and in particular they have investigated how state-based formal methods such as X-machines and its communicating counterpart could facilitate the development of correct AMAS [KHE+03]. The IMC group is in close collaboration with the Verification & Testing as well as the Computational Biology research groups of the Department of Computer Science of the University of Sheffield. An agent is perceived as an aggregation of simple behaviours described as state machines that are able to formally specify the overall behaviour of an agent [K02]. Due to the legacy of state-based models concerning testing and verification, this framework facilitates to a great extend the complete testing of the implementation with respect to the model and allows model checking techniques to be employed in order to verify that safety properties hold in the agent model.

The same group (IMC) has also developed a formal framework that includes a lightweight methodology for the development of AMAS. This includes a language for the formal description of agents, as well as the formal foundations of complete testing and model checking accompanied by several tools [EKS03]. The research is targeted towards biology-inspired agents and swarm-based systems, in which agents exhibit basic reactive behavior with limited communication [KEH+05].



Several agent frameworks have been proposed for developing intelligent software agents and multi-agent systems that are capable to perform in dynamic environments either individually or in cooperation with other agents. These frameworks and architectures enable agents to react to changes in the environment, to deliberate by performing reasoning tasks such as option selection, desire filtering, plan elaboration, means-end reasoning, and to cooperate with other agents. These modes of behavior are usually realized by specific modules that agents may trigger according to circumstances, switching their behavior between predefined discrete behavioral modes.

Considering that the existing agent architectures and frameworks result to constraining agents’ flexible behavior, researchers in the Artificial Intelligence Laboratory of the Department of Information and Communication Systems Engineering – University of the Aegean aim to devise a non-layered BDI–architecture for supporting performance in dynamic and unpredictable multi-agent environments through efficient balancing between different behavioral modes in a continuous behavioral space. This space is circumscribed by the purely (individual) reactive, the purely (individual) deliberative and the social behavioral modes. The Intelligent Collaborative Agent (ICagent) framework achieves these objectives by relating agent’s flexible behavior to cognition and sociability, supporting the management of plans constructed by the agent’s mental and domain actions in a coordinated manner. Specifically, as already pointed out in [KV01], for an agent to be flexible (i.e. to have the ability to adapt and balance between different behavioral modes) its behavioral mode cannot be determined by mere mappings of domain tasks, environment states or events to specific types of behavior: Agents may need to adapt their behavior several times in response to changes in the environment while they pursue their goals. Furthermore, although there may be some primary types of behavior, it would be desirable agents to adapt their behavior by arbitrarily “mixing” these primary types, resulting to modes that were not foreseen as such during agent or multi-agent system design. In the ICagent framework, agents adapt their behavior with respect to the way several mental actions are performed. These mental actions control the way agents perceive their environment, plan and act in it.

In a different approach, researchers from the Decision Support Systems Laboratory, Technical University of Crete, are addressing the agents’ behavioral model by adopting a well-established branch of decision theory, the multiple criteria paradigm, as a reasoning mechanism [DM07]. They have proposed a consistent modeling procedure in order to represent the relative information and to construct a preference model for agents. As above, the ultimate goal is to allow agents to perceive the environment and act rationally in it.

Aiming to support collaborative activity of humans within organized settings, members of the Artificial Intelligence Laboratory of the University of the Aegean have introduced a set of constructs for specifying organizational structures in conjunction to an explicit representation of individual and collaborative responsibilities of agents [PV06]. Special emphasis has been given to state recognition recipes that drive group members within organizations to form “acceptances”. Furthermore, agents exploit state achievement recipes for achieving goal states and fulfilling responsibilities collaboratively. To form acceptances [PV06B], agents adopt a non-summative account of group belief. These support well-organized groups to decide on their group beliefs, depending on the beliefs of their individual members and based on shared group policies. Specifically, members of the AI-Lab consider that a well-organized group believes a state s if the following conditions hold: (a) Some members of the group collaboratively decide to accept s as the view of the group, which is a view that all group members shall accept, (b) the other members of the group accept s, and (c) all the members of the group know that s has been accepted by the group: A group accepts s, or there is a group acceptance concerning s, if there is a group belief concerning s.

In an effort to address the issue of formal modelling of self-organisation, self-assembly and emergence, researchers from the IMC group at CITY College, Thessaloniki, have proposed a flexible framework, called OPERAS, which could integrate various formal methods [SKG08]. The most prominent method is inspired by Membrane Computing, a new paradigm for computation. In particular, P Systems [REFERENCE-MISSING???] provided the means to model change in a multi-agent organization. This is a characteristic of all biological and biologically–inspired systems that include colonies of social insects, flocking, schooling and in general swarm intelligence [SKG07], since agents change the way that they communicate, new agents appear in a system while others cease to exist or simply leave the system.



Decentralized control of large-scale systems of cooperative agents is a hard problem in the general case: The computation of an optimal control policy when each agent possesses an approximate partial view of the state of the environment (which is the case for large-scale systems) and agents’ observations are interdependent (i.e. one agent’s actions affect the observations of the other) is very hard even if agents’ activities are independent (i.e. the state of one agent does not depend on the activities of the other). Decentralized control of such a system in cases where agents have to act jointly is even more complex. In the case of joint activity, subsets of agents have to form teams in order to perform tasks subject to ordering constraints. Acting as a team, each group has to be coordinated by scheduling subsidiary tasks with respect to temporal and possibly other constrains, as well as other tasks that team members aim to perform (e.g. as members of other teams).

Being interested in the development of decentralized policies that can support agents to manage their resources efficiently, members of the Artificial Intelligence Lab of the University of the Aegean have proposed a method that builds on self-organization approaches for ad-hoc networks, token-based approaches for coordination in large-scale systems, and distributed constraint satisfaction [TPV+07-AAMAS 2007]. Aiming to increase the efficiency and the benefits of a multi-agent system, this method:
(a) Integrates searching, task allocation and scheduling in large-scale dynamic systems of cooperative agents. The proposed method is demonstrated by allocating temporally interdependent tasks with specific capability requirements to time-bounded agents.

(b) Demonstrates how the interplay of simple searching, task allocation and scheduling techniques using routing indices, with the dynamic self-organization of the acquaintance network to an overlay network of gateway agents, can contribute to solving this complex problem efficiently.

On the other hand, in open multi-agent systems, agents need resources provided by other agents but they are not aware of which agents provide the particular resources. Most solutions to this problem are based on a central directory that maintains a mapping between agents and resources. However, such solutions do not scale well, since the central directory becomes a bottleneck in terms of both performance and reliability. Members of the DMOD group at the University of Ioannina have proposed a fully distributed approach to this problem. In this approach, each agent maintains a limited size local cache in which it keeps information about k different resources, that is, for each of k resources, it stores the contact information of one agent that provides it. This creates a directed network of caches. The performance of three different search algorithms for navigating through this network of caches, namely, flooding, teeming, and random paths, was studied both analytically and experimentally [DP03a]. The analytical results were extended in [DP03b] for the case of skewed search requests. A suite of cache update policies that combine pull-based invalidation that is initiated by the agent that maintains the cache with push-based invalidation that is initiated by the agent that moves were proposed in [LDP04]. Experimental results indicate that a novel variation of flooding for push where a moving agent propagates its new location to agents in its old neighborhood achieves good cache consistency with a small message overhead.

In a related line of research, members of the Intelligent Systems & Robotics Laboratory of the Technical University of Crete, have proposed a gossip-based distributed algorithm, called Newscast EM, for information sharing in the task of Gaussian mixture learning [KV05]. The algorithm operates on network topologies where each node (agent) observes a local quantity and can communicate with other nodes in an arbitrary point-to-point fashion. The main difference between Newscast EM and the standard Expectation-Maximization algorithm for Gaussian mixture learning is that the M-step of the former is implemented in a decentralized manner: random pairs of agents repeatedly exchange their local parameter estimates and combine them by weighted averaging. Under such a protocol, it was shown theoretically and experimentally that nodes converge exponentially fast to the correct estimates in each step of the EM algorithm.

Concerning the task scheduling problem, researchers from the Decision Support Systems Laboratory of the Technical University of Crete have proposed:
1. The multi-agent system AgentAllocator [MD03], which is an easy to use, platform independent application, that implements a multi-criteria method to support the decision of Task Allocation. The decision maker is able to model the problem (according to his policy) through its inputs dialogs and employ the final solution proposed by the system.

2. A Multi-criteria Protocol for Multi-agent Negotiation [MD04]. Negotiation processes are often characterized by conflicts of interests of the negotiating parts. However, it is possible to mitigate these conflicts if we support the negotiation process with a well-structured model. This area of interest has large occupied the scientists of Group Decision Support Systems (GDSS) and particularly those who focus on the Negotiation Support Systems (NSS). The proposed system involves an experimental multi-criteria prototype negotiation protocol that allows agents to follow a process in order to end up with an optimal decision. The proposed model is able to estimate agents’ preferences and suggest convenient solutions.
3. A spectral clustering approach to designate an efficient scheduling plan [DDM08]. The method’s contribution is twofold: (a) during a system’s runtime, the method allows agents to be coordinated in a highly-demanding workflow process; (b) during a system’s design time, it specifies the amount of the resources needed, preventing a wasteful architecture.



In the survey paper [K03], Koubarakis argues that P2P system and MAS have much to gain for each other, and that combinations of techniques from these two areas can lead to important advances in Internet Computing applications. The integration of P2P and MAS is also the focus of the annual International Workshop on Agents and Peer-to-Peer Computing ( since 2002.

Researchers from the Department of Informatics and Telecommunications, National and Kapodistrian University of Athens, have demonstrated a successful integration of techniques from MAS and P2P by presenting P2P-DIET, an extensible P2P service implemented using the DIET agents platform [MK05, HWB+02,] This platform draws inspiration from nature, where many living organisms interact in diverse ways so as produce complex ecosystems. The platform is designed around agents with minimal properties and limited individual capabilities. DIET agents are not assumed to be highly intelligent or to use complex communication protocols. Intelligent behaviour can emerge from the interaction between agents.

The DIET Agents platform is designed in a three-layer architecture. The lowest layer, the core layer, contains the DIET kernel, enabling DIET environments, providing the basic capabilities for agent creation. The kernel also provides for connections between agents, which allows messages to be passed between them. The core layer also contains basic support for debugging and visualisation. The application reusable component layer contains software components that are reusable between multiple applications, but are not essential for inclusion in the core layer. Examples of the components included are ones supporting remote communication and event scheduling. The application layer is the top layer. It contains code specific to particular applications, as well as debugging and visualisation code that may be application specific.

A direct descendant of DIAS is the system P2P-DIET [IK04, IKT04a, IKT04b, KTI+03,]. This is a Distributed Information Alert System for digital libraries, which was presented in [KKT+02,KTR+02,KT02] but was not fully implemented. P2P-DIET is a super-peer system and has two kinds of nodes: super-peers and clients (see Fig. 1). Super-peers are equal and have the same responsibilities. Each super-peer serves a fraction of the clients and keeps indices on the resources of those clients to be able to answer queries efficiently. Clients can run on user computers where resources are also stored. Clients interact directly with one another to access resources.


Fig. 1. The P2P-DIET System


P2P-DIET supports the typical one-time query scenario of P2P networks. A client can send a query (e.g., “I want music by Moby”) to its access point and the access point will broadcast this query to all super-peers. In this way, answers will be produced for all matching network resources. Answers are returned to the access point of the client originating the query and are then passed to the client for further processing. P2P-DIET also supports long-standing (continuous) query scenarios (e.g., “Notify me when a song of Moby becomes available”). Clients may subscribe to the system with a continuous query expressing their long-standing information needs. Whenever a resource is published at an access point, P2P-DIET makes sure that clients with profiles matching the metadata of this resource are notified. A client can be connected to P2P-DIET through a single super-peer node, which is the access point of the client. Clients are allowed to migrate to a different access point and can use dynamic IP addresses. Clients can connect, disconnect or even leave the system silently at any time. When a client is off-line, notifications matching its continuous queries are stored by the access point of the client and are delivered to it the next time that it connects to the network. A similar situation is when a client A requests a resource, but the resource owner client B is not on-line. In this case, the client A may request a rendezvous with the resource. When client B later-on reconnects to the network, its access point informs it that the resource must be delivered to the access point of the client. P2P-DIET provides message authentication and message encryption using public key cryptography. Public/private keys are also used to securely identify peers since it is not possible to identify a peer from its IP address because peers may use dynamic IP addresses.

Nodes in P2P-DIET are implemented as DIET environments where different types of agents live. The P2P-DIET implementation makes use of the capabilities of lightweight mobile agents offered by the DIET kernel to implement various local management tasks and P2P protocols. For example, in each super-peer environment, a data management agent keeps indices on resource meta-data and continuous queries to achieve scalable query processing and filtering by each super-peer. In each super-peer environment, a router agent achieves correct flow of network messages by using shortest paths and minimum-weight spanning trees. For each ad-hoc query posed by a user, a query answering agent starts from the user’s client environment and, using information from router agents, migrates to all super-peers to find all resources that match the user query. Similarly, for each continuous query subscribed by a user, a subscriber agent starts from the user’s client environment and migrates to appropriate super-peer environments to subscribe the query. Migration is performed towards all super-peers that might end-up with resource metadata matching the query and is only constrained by taking into account query subsumption relations. Subscription migration has the effect that filtering takes place closer to the clients posting the resources to curtail message propagation in the network.

P2P-DIET was built with the intention to show that the DIET Agents platform with its minimal agents and self-organization capabilities can be used to develop large scale P2P applications. The P2P-DIET application takes advantage of self-organisation at two levels: the super-peer network and the DIET Agents platform that is used to realize P2P-DIET. At the first level, P2P-DIET super-peers self-organize into a network which deals efficiently with pull and push information requests while, at the same time, adapting to super-peer joins, leaves or failures. At the second level, lightweight DIET Agents self-organize themselves in order to implement P2P-DIET functionalities. For example, query-answering agents begin from a client node and migrate to remote nodes with the purpose of answering a one-time query. At each node that they arrive, they interact with router agents and data management agents to decide their next destination node or to compute partial query answers. The next operation to be performed by an agent is not determined by any kind of higher level of control. On the contrary, its current status and the state of its environment determine agent decisions. For example, an agent might choose to replicate itself if there are more than one possible route to follow for answering a query. Newly created agents are totally independent to travel around the network and collect query answers. P2P-DIET is an extensible system for the development of applications in need of the two scenarios discussed above (see the layered view of the system on the righthand side of Fig. 1). In the current demo of the system publications and subscriptions are expressed using a well-understood attribute-value model called AWPS in [KTK+02]. AWPS is based on named attributes with value text interpreted under the Boolean and VSM or LSI models. The query language of AWPS allows Boolean combinations of comparisons A op v, where A is an attribute, v is a text value and op is one of the operators “equals”, “contains” or “similar” (“equals” and “contains” are Boolean operators and “similar” is interpreted using the VSM or LSI model of Information Retrieval. [TKD04] provides detailed performance analysis of some of the indexing algorithms for AWPS used in P2P-DIET and demonstrates very good efficiency and scalability properties (e.g., upon receiving a new publication, a P2P-DIET node can filter 3 millions of long-standing queries in just under 200 milliseconds).



Adaptation and Learning has a long tradition and has received much attention due to the necessity of managing self-organizing properties of complex systems composed of autonomous entities. Of particular importance are adaptation and learning in multi-agent systems contexts.

In the ICagent framework developed by members of the Artificial Intelligence Laboratory of the University of the Aegean, agents adapt their behavior with respect to the way several mental actions are performed. These mental actions control the way agents perceive their environment, plan and act in it. Therefore, the type of behavior adopted by an agent at a specific time point is considered as a property that emerges as the agent performs, according to the perceived state of the environment, to the occurring events, and to the agent’s mental state. Adapting their behavior, agents:
1. Decide about which facts and events they must monitor in their physical environment, as well as the way of doing this.
2. Decide whether they shall reason about the relative strength of their desires and intentions, or whether they will commit to fulfill a desire without considering other intentions they already have.

3. Decide whether they will assess their options towards fulfilling a desire or, acting purely reactively, whether they will fetch a good solution and start pursuing it.
4. Determine whether and how they shall generate and elaborate their plans individually or jointly with other agents.

Work on adaptation of multi-agent systems by members of the same Laboratory (AI-LAB) concerns self-tuning large-scale networks of agents for effective and efficient information searching and sharing. Specifically, the problem here is to find the right agents in a large and dynamic network to provide the needed resources in a timely fashion. The method proposed in [V07] is a method for information searching and sharing that combines routing indices with token-based methods. The proposed method enables agents to search effectively by acquiring their neighbors’ interests, advertising their information provision abilities and maintaining indices for routing queries, in an integrated way. Specifically, [ISSN – AAMAS 2007] demonstrates through performance experiments how static and dynamic networks of agents can be ‘tuned’ to answer queries effectively as they gather evidence for the interests and information provision abilities of others, without altering the topology or imposing an overlay structure to the network of acquaintances. An enhancement of this method has been presented in [V08] considering agents that do not share a conceptualization of their domains (heterogeneous agents). This later method concerns the assertion of logical shortcuts between peers, resulting to the emergence of overlay structures.

In a different line of research, members of the Intelligent Systems & Robotics Laboratory of the Technical University of Crete, have developed a set of scalable Reinforcement Learning algorithms for learning the behavior of a group of agents in a collaborative multi-agent setting [KV06]. The proposed framework exploits the dependencies between agents by decomposing the global team payoff function into a sum of local terms, forming a factor graph. On this graph they apply a payoff propagation algorithm (analogous to belief propagation for Bayesian networks) that computes near-optimal individual actions for the agents. Based on the same framework, they have also proposed a suite of model-free reinforcement learning techniques, called Sparse Cooperative Q-learning, which approximate the global action-value function of the team, as defined on the factor graph, and perform updates using the contribution of the individual agents to the maximal global action value. This results in a fast and easy-to-implement algorithm that scales only linearly in the problem size.



Unknown environments are unpredictable, inherently inaccessible, non-deterministic, and often dynamic. The fact that agents have bounded resources has important consequences to agent’s performance in such environments (with respect to their perception, deliberation, performance, and cooperation abilities). First, agents cannot be omniscient; therefore, they cannot be fully aware of the changes that occur in their physical environment. Continuous monitoring of the environment and constant communication with peers are costly tasks. Furthermore, the detection of new, irrelevant to agents’ interests events may indicate new problems, or suggest new opportunities that lead to redundant deliberation. Therefore, agents must be able to focus only to those events that are relevant to their endeavor to achieve their goals, or to events that are significant for initiating new commitments. The time that agents have in order to compute responses is always limited and bounded to the time that they have until their resources exceed. Thus, the time the agents have in their disposal to fulfill their goals is always limited and bounded by the constraints that the physical environment, their capabilities, and their goals impose. In addition to time constrains, agents have to plan and realize their actions so as to achieve their goals without exceeding environmental and own resources. Thus, agents often need to work together. Collaboration between two or more agents may be initiated either when an agent has not the capability to achieve a goal alone, or when it believes that by performing jointly with other agents it can achieve better results (e.g. it can save resources, or increase its expected utility).

Supporting a mental-state view of plans, members of the Artificial Intelligence Lab of the Department of Information and Communication Systems Engineering of the University of the Aegean, point to considering the following issues [ICAGENT]: (a) How the different types of behavior affect and are affected by agents’ mental state? (b) What types of changes in agent’s mental state and how these changes drive the adaptation of agents’ behavior? (c) How these changes are incorporated while the agent plans and acts, and how adaptation is achieved? With the aim to provide answers to these questions, members of this Laboratory (AI-LAB) present the ICagent framework for agent development. This framework provides:
1. A mental-state view of plans: Agents have plans when they have a particular set of beliefs and intentions.
2. Reasoning tasks for agents to manage their plans towards adapting and balancing their behavior between reactive, deliberative and social.
3. Mental actions for performing the reasoning tasks: These actions are treated uniformly and in coordination with the domain actions. Therefore, their performance can affect agents' overall behavioral mode according to agents beliefs and intentions.

4. A generic framework for agents’ collaboration (social behavior) that is based on the SharedPlans model of collaborative activity.

Adopting a decision-theoretic approach, members of the Intelligent Systems & Robotics Laboratory of the Technical University of Crete, have developed scalable algorithms for solving Decentralized partially observable Markov decision processes (Dec-POMDPs) [OSW+08]. Dec-POMDPs constitute an expressive framework for multi-agent planning under uncertainty, but solving them is provably intractable. In this work, the authors have demonstrated how to improve the scalability of approximate solution algorithms by exploiting locality of interaction between agents in a factored representation. Factored Dec-POMDP representations have been proposed before, but only for Dec-POMDPs whose transition and observation models are fully independent. Such strong assumptions simplify the planning problem, but result in models with limited applicability. By contrast, in the above work the authors consider general factored Dec-POMDPs for which they analyze the model dependencies over space (locality of interaction) and time (horizon of the problem). They also present a formulation of decomposable value functions. Together, their results show how to exploit the problem structure as well as heuristics in a single framework that is based on collaborative graphical Bayesian games. Preliminary experiments show speedups of about two orders of magnitude over other methods.



Part of the mobile computing research performed in the DMOD group at the University of Ioannina explored the potential use of mobile agent frameworks in designing more flexible and extensible database architectures [PCS-toappear]. Mobile agents have been used for developing a new approach to accessing relational database systems that is especially appropriate for thin clients and wireless communications [PSP00]. This approach was extended towards the dynamic materialization of views over multiple database management systems [KSCP04]. In a similar line of research, mobile agent frameworks were used for designing and building middleware services for mobile users [SSPE04]. One such middleware service is the locker service that allows users of mobile devices to rent storage at the fixed network [VIP02].

In a different application domain, researchers from the Decision Support Systems Laboratory of the Technical University of Crete have proposed an agent-based system implementing an original consumer-based methodology for product penetration strategy selection in real world situations [MMP+03]. In this system, agents are simultaneously considered according to two different levels, a functional and a structural level. In the functional level there are three types of agents: task agents, information agents, and interface agents assuming task’s fulfilment through cooperation, information gathering tasks, and mediation between users and artificial agents respectively. In the structural level there are elementary agents based on a generic reusable architecture, and complex agents created dynamically in a hierarchical way.



As it is pointed in the introduction of this chapter, research concerning agent technologies can be viewed from three perspectives: As a way for engineering applications around autonomous communicative components, as media for representing complex and dynamic real-world environments, and as a source of new technologies related to the abilities of agents. These goals align with the goals of the Greek community whose members, as described above, are quite active in this field of artificial intelligence, with important results for the AMAS community. Although much more to be done, this will be an important and exciting topic for many of us, requiring new brains and collaborative activities.


[ASP07] A. Artikis, M. Sergot, and J. Pitt, An Executable Specification of a Formal Argumentation Protocol, Artificial Intelligence Journal, 171(10-15):776-804, 2007.

[ASP08] A. Artikis, M. Sergot, and J. Pitt, Specifying Norm-Governed Computational Societies, ACM Transactions on Computational Logic, to appear in 2008.

[APS02] A. Artikis, J. Pitt, and M. Sergot, Animated Specifications of Computational Societies, in the Proceedings of Autonomous Agents and Multi-Agent Systems (AAMAS), pp. 1053-1062, Bologna, 2002.

[DM07] P. Delias and N.F. Matsatsinis. The multiple criteria paradigm as a background for agent methodologies. In 8th Annual International Workshop "Engineering Societies in the Agents World", Athens, Greece, 2007.

[DDM08] P. Delias, A. Doulamis, and N. Matsatsinis, A Joint Optimization Algorithm for Dispatching Tasks in Agent-based Workflow Management Systems, in 10th International Conference on Enterprise Information Systems, ICEIS, Barcelona, Spain, 2008.

[DP03a] V. V. Dimakopoulos and E. Pitoura, Performance Analysis of Distributed Search in Open Agent Systems, International Parallel and Distributed Processing Symposium (IPDPS), Nice, France, April 2003.

[DP03b] V. V. Dimakopoulos and E. Pitoura, A Peer-to-Peer Approach to Resource Discovery in Multi-Agent Systems. CIA 2003, Helsinki, Finland, 2003.

[EKS03] G. Eleftherakis, P. Kefalas, A. Sotiriadou, Formal Verification of Reactive Agents for Intelligent Control, In Proceedings of the 12 ISAP Intelligent System Applications to Power Systems Conference, 2003.

[HWB+02] C. Hoile, F. Wang, E. Bonsma and P. Marrow, Core Specification and Experiments in DIET: A Decentralised Ecosystem-inspired Mobile Agent System" Proc. 1st Int. Conf. on Autonomous Agents and Multi-Agent Systems (AAMAS2002), pp. 623-630, Bologna, Italy, 2002.

[IK04] S. Idreos and M. Koubarakis, P2P-DIET: Ad-hoc and Continuous Queries in Peer-to-Peer Networks using Mobile Agents, 3rd Hellenic Conference in Artificial Intelligence, Samos, Greece, May 5-8, 2004. In LNAI, Vol. 3025, pages 23-32.

[IKT04a] S. Idreos, M. Koubarakis and C. Tryfonopoulos. P2P-DIET: One-Time and Continuous Queries in Super-peer Networks. Proceedings of the IX International Conference on Extending Database Technology (EDBT04), Heraklion, Crete, Greece, March 14-18, 2004. In LNCS, Vol. 2992, pages 851-853.

[IKT04b] S. Idreos, M. Koubarakis and C. Tryfonopoulos, P2P-DIET: An Extensible P2P Service that Unifies Ad-hoc and Continuous Querying in Super-peer Networks, Proceedings of the ACM SIGMOD/PODS 2004 Conference. Maison de la Chimie, Paris, France, June 13-18, 2004.

[KP07] D. Kaponis and J. Pitt. Dynamic specifications in norm-governed open computational societies. In Proceedings of Workshop on Engineering Societies in the Agents’ World (ESAW). Springer, 2007.

[KSCP04] K. Karenos, G. Samaras, P. K. Chrysanthis, E. Pitoura, Mobile Agent-Based Services for View Materialization. Mobile Computing and Communications Review 8(3): 32-43 2004.

[KEH+05] P. Kefalas, G. Eleftherakis, M. Holcombe, I. Stamatopoulou, "Formal Modelling of the Dynamic Behaviour of Biology-Inspired Agent-based Systems", In Molecular Computational Models: Unconventional Approaches, M. Gheorghe (ed.), Idea Group Publishing, pp. 243-276, 2005.

[KHE+03] P. Kefalas, M. Holcombe, G. Eleftherakis, M. Gheorge, "A Formal Method for the Development of Agent Based Systems", In Intelligent Agent Software Engineering, V.Plekhanova (eds), Idea Group Publishing Co., pp.68-98, 2003

[K02] P. Kefalas, "Formal Modelling of Reactive Agents as an Aggregation of Simple Behaviours", Methods and Applications of Artificial Intelligence, Lecture Notes in Artificial Intelligence 2308, I.P.Vlahavas and C.D.Spyropoulos (Eds.), Springer, pp.461-472, 2002.

[KV06] J.R. Kok and N. Vlassis. Collaborative multiagent reinforcement learning by payoff propagation. Journal of Machine Learning Research, 7:1789-1828, 2006.

[KTI+03] M. Koubarakis, C. Tryfonopoulos, S. Idreos and Y. Drougas, Selective Information Dissemination in P2P Networks: Problems and Solutions, ACM SIGMOD Record, Special issue on Peer-to-Peer Data Management, Karl Aberer (editor), Volume 32, Number 3, September 2003.

[K03] M. Koubarakis, Multi-Agent Systems and P2P Computing: Methods, Systems and Challenges (Invited talk), Proceedings of the 7th International Workshop on Cooperative Information Agents (CIA 2003), Helsinki, Finland, August 27-29, 2003. In LNAI, Vol. 2782, pages 46-61, Springer.

[KT02] M. Koubarakis and C. Tryfonopoulos, Peer-to-peer agent systems for textual information dissemination: algorithms and complexity. In UK Workshop on Multiagent Systems (UKMAS-2002), Liverpool, UK, 18 & 19 December, 2002.

[KTR+02] M. Koubarakis, C. Tryfonopoulos, P. Raftopoulou, T. Koutris, Data Models and Languages for Agent-Based Textual Information Dissemination. 6th International Workshop on Cooperative Information Systems (CIA 02), 18-20 September 2002, Universidad Rey Juan Carlos, Madrid, Spain.

[KKT+02] M. Koubarakis, T. Koutris, C. Tryfonopoulos, P. Raftopoulou, Information Alert in Distributed Digital Libraries: The Models, Languages and Architecture of DIAS. 6th European Conference on Research and Advanced Technology for Digital Libraries (ECDL 02), 16-18 September 2002, Pontifical Gregorian University, Rome, Italy.

[KV06] V. Kourakos-Mavromichalis, G. Vouros, Building Intelligent Collaborative Interface Agents with the ICagent Development Framework. International Journal of Agents and Multi-Agent Systems, Vol 13, Issue 2, 2006.

[KV06B] V. Kourakos-Mavromichalis and G. Vouros. Behaviour Flexibility in Dynamic and Unpredictable Environments: The ICAGENT Approach. To appear in: Proceedings of 4th Hellenic Conference on Artificial Intelligence, Lecture Notes in Artificial Intelligence (LNAI), Springer Vol 3955, G.Antoniou et al (Eds), 2006.

[KV05] W. Kowalczyk and N. Vlassis. Newscast EM. In L. K. Saul, Y. Weiss, and L. Bottou, editors, Advances in Neural Information Processing Systems (NIPS) 17, pages 713-720. MIT Press, Cambridge, MA, 2005.

[LDP04] E. Leontiadis, V. V. Dimakopoulos and E. Pitoura, Cache Updates in a Peer-to-Peer Network of Mobile Agents. 4th IEEE International Conference on Peer-to-Peer Computing (P2P 2004), Zurich, Switzerland, August 2004.

[LMS+05] M.Luck, P. McBurney, O. Shehory, S. Willmott and the Agent Link Community. “Agent Technology: Computing as Interaction. A Roadmap for Agent Based Computing”. AgentLink III, September 2005.

[MKL+01] P. Marrow, M. Koubarakis, R.H. van Lengen, F. Valverde-Albacete, E. Bonsma, J. Cid-Suerio, A.R. Figueiras-Vidal, A. Gallardo-Antolin, C. Hoile, T. Koutris, H. Molina-Bulla, A. Navia-Vazquez, P. Raftopoulou, N. Skarmeas, C. Tryfonopoulos, F. Wang, C. Xiruhaki, Agents in Decentralised Information Ecosystems: The DIET Approach. Symposium on Information Agents for E-Commerce, AISB'01 Convention, 21st - 24th March 2001 University of York, United Kingdom.

[MK05] P. Marrow and M. Koubarakis, Self-organising applications using lightweight agents. Third International Workshop on Engineering Self-Organising Applications (ESOA 2005). Held in conjunction with the 4th International Joint Conference on Autonomous Agents and Multi-Agent Systems (AAMAS), Utrecht, The Netherlands, 2005.

[MMP+03] N.F. Matsatsinis, P. Moraϊtis, V. Psomatakis, and N. Spanoudakis, An Agent-Based System for Products Penetration Strategy Selection, Applied Artificial Intelligence: An International Journal, vol. 17, no. 10, pp. 901-925, 2003.

[MD03] N.F. Matsatsinis and P. Delias, AgentAllocator: An Agent-Based Multi-criteria Decision Support System for Task Allocation, in: V. Marik, D. McFarlane, P. Valckenaers (eds.), Holonic and Multi-agent Systems for Manufacturing, Lectures Notes in Artificial Intelligence, vol. 2744, Springer-Verlag Berlin Heidelberg, pp. 225-235, 2003.

[MD04] N.F. Matsatsinis and P. Delias, A Multi-criteria Protocol for Multi-agent Negotiations, in: G.A. Vouros and T. Panayiotopoulos (Eds.), Methods and Applications of Artificial Intelligence, Lectures Notes in Artificial Intelligence, vol. 3025, Springer-Verlag Berlin Heidelberg, pp. 103–111, 2004.

[OSW+08] F. A. Oliehoek, M. T. J. Spaan, S. Whiteson, and N. Vlassis. Exploiting Locality of Interaction in Factored Dec-POMDPs. In Proc. Int. Joint Conf. on Autonomous Agents and Multi-Agent Systems, Estoril, Portugal, May 2008.

[PSP00] S. Papastavrou, G. Samaras and E. Pitoura, Mobile Agents for Distributed WWW Access, IEEE Transactions on Knowledge and Data Engineering, 12(5), pp 802-820, 2000.

[PV06] I. Partsakoulakis and G. Vouros, Agent-Enhanced Collaborative Activity in Organized Settings. International Journal of Cooperative Information Systems, 15(1), March 2006.

[PV06B] I. Partsakoulakis and G. Vouros, Building Common Awareness in Agent Organizations. International Journal of Knowledge-Based and Intelligent Engineering (KES), special edition on Agent-Mediated Knowledge Management, 10(4), IOS Press, 2006.

[PV04] I. Partsakoulakis, G. Vouros, Roles in MAS: Managing the Complexity of Tasks and Environments. In "Multi-Agent Systems: An Application Science" T.Wagner (Editor), Kluwer Academic, 2004.

[PCS-toappear] E. Pitoura, P. Chrysanthis, and G. Samaras, Distributed Databases and Transaction Processing. In Mobile Agents in Networking and Distributed Computing, Jiannong Cao and Sajal Das, eds, John Wiley, To appear.

[SSP+04] C. Spyrou, G. Samaras, E. Pitoura and P. Evripidou, Mobile Agents for Wireless Computing: The Convergence of Wireless Computation Models with Mobile Agent Technologies. ACM/Baltzer MONET, Volume 9, No 5, Oct 2004.

[SKG08] I. Stamatopoulou, P. Kefalas, and M. Gheorghe. OPERAS: a formal framework for multi-agent systems and its application to swarm-based systems. LNAI, A.Artikis, G.O'Hare, K.Stathis, and G.Vouros (Eds.), (to be published), Springer, 2008.

[SKG07] I. Stamatopoulou, P. Kefalas, and M. Gheorghe. Modelling the dynamic structure of biological state-based systems. BioSystems, 87(2-3):142-149, February 2007.

[TPV+07] C. Theocharopoulou, I. Partsakoulakis, G. Vouros, C. Stergiou, Overlay networks for task allocation and coordination in dynamic large-scale networks of cooperative, AAMAS 2007, Honolulu, Hawaii, 2007.

[TKD04] C. Tryfonopoulos and M. Koubarakis and Y. Drougas, Filtering Algorithms for Information Retrieval Models with Named Attributes and Proximity Operators, Proceedings of the 27th Annual ACM SIGIR Conference. July 25-July 29, 2004, Sheffield, United Kingdom.

[VIP02] Y. Villate, A. Illarramendi and E. Pitoura, Keep Your Data Safe and Available While Roaming. International Journal of Mobile Networks and Application (MONET), Special Issue on Pervasive Computing, 7(4): 315-328, August 2002.

[V07] G.Vouros, Information Searching  and Sharing in Large-Scale Dynamic Networks, AAMAS 2007, Honolulu, Hawaii, 2007.

[V08] G.Vouros, Searching and Sharing Information in Networks of Heterogeneous Agents, poster in AAMAS 2008.