Skip to main content



Invited Talks

Tournament Solutions and Their Applications to Multiagent Decision Making

Given a finite set of alternatives and choices between all pairs of alternatives, how to choose from the entire set in a way that is faithful to the pairwise comparisons? This simple, yet captivating, problem is studied in the literature on tournament solutions. A tournament solution thus seeks to identify the “best” elements according to some binary dominance relation, which is usually assumed to be asymmetric and complete. As the ordinary notion of maximality may return no elements due to cyclical dominations, numerous alternative solution concepts have been devised and axiomatized.
Many problems in multiagent decision making can be addressed using tournament solutions. For instance, tournament solutions play an important role in collective decision-making (social choice theory), where the binary relation is typically defined via pairwise majority voting. Other application areas include adversarial decision-making (theory of zero-sum games) and coalitional decision-making (cooperative game theory) as well as multi-criteria decision analysis and argumentation theory.
In this talk, I will present an overview of some of the most common tournament solutions such as the uncovered set, the minimal covering set, and the bipartisan set and analyze them from an algorithmic point of view.
Felix Brandt

Research Challenges in Simulation Aided Design of Complex Multi-agent Systems

In today’s world, we are increasingly surrounded by and reliant on complex systems and infrastructures. Often, these systems behave far from the optimum or even highly undesirable. Roads in our cities are congested, plane trips frequently delayed, computer networks routinely overrun by worms and electricity grids fail in split-second cascade reactions. Our systems have become massively interwoven and interdependent making both highly positive and negative chain reactions possible in critical systems.
Michal Pěchouček, Michal Jakob

Models and Specifications

A Model Driven Development of Platform-Neutral Agents

The automatic transformation of software agent designs into implementations for different agent platforms is currently a key issue in the MAS development process. Recently several approaches have been proposed using model driven development concepts to specify generic agent metamodels and/or define a set of transformation rules from the design phase for different agent implementation platforms. Although for some systems this is acceptable, in the context of Ambient Intelligence, this could be a serious limitation because of the variety of devices involved in these systems ranging from desktop computers to lightweight devices. In this paper we propose to transform PIM4Agents, a generic agent metamodel used at the design phase, into Malaca, an agent specific platform-neutral metamodel for agents. With only one set of transformations it is possible to generate a partial implementation in Malaca, which can be deployed in any kind of device and can interact with any FIPA compliant agent platform.
Inmaculada Ayala, Mercedes Amor, Lidia Fuentes

A Novel Formal Specification Approach for Real Time Multi-Agent System Functional Requirements

A novel formal functional requirements specification approach for real-time multi-agent system is presented in this paper. The methodology of our approach consists in translating extended AUML diagrams describing RT-MAS’ functional requirements into a RT-Maude specification. The proposed approach considers jointly functional, static and dynamic aspects of real-time multi-agent systems. The functional aspects are described by a temporal AUML use case diagram and the static aspects are represented using a temporal AUML class diagram. Whereas the dynamic aspects are described using state chart (individual behavior) and an extended AUML protocol (collective behavior) diagrams. The aims of this approach are, on the one hand, to combine the advantages of the graphical modeling formalism Agent UML and the formal specification language RT-Maude in a single technique, and, on the other hand, to integrating the formal validation of the consistency of the models, since the analysis phase. The approach is illustrated using a concrete example.
Mohamed Amin Laouadi, Farid Mokhati, Hassina Seridi-Bouchelaghem

Do You Get It? User-Evaluated Explainable BDI Agents

In this paper we focus on explaining to humans the behavior of autonomous agents, i.e., explainable agents. Explainable agents are useful for many reasons including scenario-based training (e.g. disaster training), tutor and pedagogical systems, agent development and debugging, gaming, and interactive storytelling. As the aim is to generate for humans plausible and insightful explanations, user evaluation of different explanations is essential. In this paper we test the hypothesis that different explanation types are needed to explain different types of actions. We present three different, generically applicable, algorithms that automatically generate different types of explanations for actions of BDI-based agents. Quantitative analysis of a user experiment (n=30), in which users rated the usefulness and naturalness of each explanation type for different agent actions, supports our hypothesis. In addition, we present feedback from the users about how they would explain the actions themselves. Finally, we hypothesize guidelines relevant for the development of explainable BDI agents.
Joost Broekens, Maaike Harbers, Koen Hindriks, Karel van den Bosch, Catholijn Jonker, John-Jules Meyer

Trust, Norms and Reputation

Reputation in Multi Agent Systems and the Incentives to Provide Feedback

The emergence of the Internet leads to a vast increase in the number of interactions between parties that are completely alien to each other. In general, such transactions are likely to be subject to fraud and cheating. If such systems use rational software agents to negotiate and execute transactions, mechanisms that lead to favorable outcomes for all parties instead of giving rise to defective behavior are necessary to make the system work: trust and reputation mechanisms. This paper analyzes different incentive mechanisms helping these trust and reputation mechanisms in eliciting users to report own experiences honestly.
Miriam Heitz, Stefan König, Torsten Eymann

Normative Deliberation in Graded BDI Agents

Norms have been employed as a coordination mechanism for Open MAS, but to become effective, they must be internalized by agents; i.e. these agents must be able to accept norms while maintaining their autonomy. Nevertheless, traditional BDI agent architectures only represent beliefs, intentions and desires. In this paper, the multi-context BDI agent architecture has been extended with a recognition context and a normative context in order to allow agents to acquire norms from their environment and consider norms in their decisions.
Natalia Criado, Estefania Argente, Vicent Botti

Inducing Desirable Behaviour through an Incentives Infrastructure

In open multiagent systems, where agents may join/leave the system at runtime, participants can be heterogeneous, self-interested and may have been built with different architectures and languages. Therefore, in such a type of systems, we cannot assure that agents populating them will behave according to the objectives of the system. To address this problem, organisational abstractions, such as roles and norms, have been proposed as a promising solution. Norms are often coupled with penalties and rewards to deter agents from violating the rules of the system. But, what happens if a current population of agents does not care about these penalties/rewards. To deal with this problem, we propose an incentives infrastructure that allows to estimate agents’ preferences, and can modify the consequences of actions in a way that agents have incentives to act in a certain manner. Employing this infrastructure, a desirable behaviour can be induced in the agents to fulfil the preferences of the system.
Roberto Centeno, Holger Billhardt, Sascha Ossowski

Models, Tools and Architectures

SONAR/OREDI: A Tool for Creation and Deployment of Organisation Models

The need for handling the increasing complexity in software systems has allowed the introduction and establishment of an organisational paradigm as an alternative in software modelling and development. Especially within the multi-agent systems community, organisational concepts are enjoying increasing popularity for efficiently structuring multi-agent behaviour. Organisational specifications and their implementation as multi-agent systems lack however a streamlined transition between each other. In this paper we address this problem by introducing a software tool capable of creating and editing organisation models as well as deploying such models as multi-agent systems. The tool is built on Sonar, a formal organisational specification based on Petri nets. By unifying in one tool the organisational specification and deployment process quick reaction cycles to incremental changes of system design become possible.
Endri Deliu, Michael Köhler-Bußmeier

Enhancing the Interoperability between Multiagent Systems and Service-Oriented Architectures through a Model-Driven Approach

Service-orientation has become the leading paradigm for modern IT system design and development as service-oriented system design has great potential for improving the efficiency and quality of the IT systems. This paper presents a model-driven approach for the generic integration of service-oriented architectures (SOA) and multi-agent systems (MAS). In fact, a model transformation from SoaML—a metamodel for SOA—to Pim4Agents—a platform independent metamodel for MAS—is utilized for integration. The relevance of this approach is proven by applying it to a real-world industry scenario.
Christian Hahn, Sven Jacobi, David Raber

Unifying Agent and Component Concepts

Jadex Active Components
The construction of distributed applications is a challenging task due to inherent system properties like message passing and concurrency. Current technology trends further increase the necessity for novel software concepts that help dealing with these issues. An analysis of existing software paradigms has revealed that each of them has its specific strengths and weaknesses but none fits all the needs. On basis of this evaluation in this paper a new approach called active components is proposed. Active components are a consolidation of the agent paradigm, combining it with advantageous concepts of other types of software components. Active components, like agents, are autonomous with respect to their execution. Like software components, they are managed entities, which exhibit clear interfaces making their functionality explicit. The approach considerably broadens the scope of applications that can be built as heterogeneous component types, e.g. agents and workflows, can be used in the same application without interoperability problems and with a shared toolset at hand for development, runtime monitoring and debugging. The paper devises main characteristics of active components and highlights a system architecture and its implementation in the Jadex Active Component infrastructure. The usefulness of the approach is further explained with an example use case, which shows how a workflow management system can be built on top of the existing infrastructure.
Alexander Pokahr, Lars Braubach, Kai Jander

Applications I

Impact of Competition on Quality of Service in Demand Responsive Transit

Demand responsive transportation has the potential to provide efficient public door-to-door transport with a high quality. In currently implemented systems in the Netherlands, however, we observe a decrease in the quality of service (QoS), expressed in longer travel times for the customers. Currently, generally one transport company is responsible for transporting all customers located in a specified geographic zone. In general it is known that when multiple companies compete on costs, the price for customers decreases. In this paper, we investigate whether a similar result can be achieved when competing on quality instead. To arrive at some first conclusions, we set up a multiagent environment to simulate the assignment of rides to companies through an auction on QoS, and the insertion of allocated rides in the companies’ schedules using online optimization. Our results reveal that this set-up improves the quality of the service offered to the customers at moderately higher costs.
Ferdi Grootenboers, Mathijs de Weerdt, Mahdi Zargayouna

Towards Distributed Agent Environments for Pervasive Healthcare

In this paper we present a prototypical pervasive health care infrastructure, whose purpose is the continuous monitoring of pregnant women with gestational diabetes mellitus. In this infrastructure, patients are equipped with a body-area network made of sensors to control blood pressure and glucose levels, where the sensors are connected to a smart phone working as a hub to collect the data. These data is then fed to a pervasive GRID where abductive agents provide a diagnosis for the actual reading of the sensors and contacting health care professionals if necessary. We also show how, by applying the concept of agent environment, we are facilitated in defining a pervasive GRID for roaming agents that monitor continuously the health status of the patients.
Stefano Bromuri, Michael Ignaz Schumacher, Kostas Stathis

Context-Aware Route Planning

In context-aware route planning, there is a set of transportation agents each with a start and destination location on a shared infrastructure. Each agent wants to find a shortest-time route plan without colliding with any of the other agents, or ending up in a deadlock situation. We present a single-agent route planning algorithm that is both optimal and conflict-free. We also present a set of experiments that compare our algorithm to finding a conflict-free schedule along a fixed path. In particular, we will compare our algorithm to the approach where the shortest conflict-free schedule is chosen along one of k shortest paths. Although neither approach can guarantee optimality with regard to the total set of agent route plans — and indeed examples can be constructed to show that either approach can outperform the other — our experiments show that our approach consistently outperforms fixed-path scheduling.
Adriaan W. ter Mors, Cees Witteveen, Jonne Zutt, Fernando A. Kuipers

Coordination and Learning

Social Conformity and Its Convergence for Reinforcement Learning

A dynamic environment whose behavior may change in time presents a challenge that agents located there will have to solve. Changes in an environment e.g. a market, can be quite drastic: from changing the dependencies of some products to add new actions to build new products. The agents working in this environment would have to be ready to embrace this changes to improve their performance which otherwise would be diminished. Also, they should try to cooperate or compete against others, when appropriated, to reach their goals faster than in an individual fashion, showing an always desirable emergent behavior. In this paper a reinforcement learning method proposal, guided by social interaction between agents, is presented. The proposal aims to show that adaptation is performed independently by the society, without explicitly reporting that changes have occurred by a central authority, or even by trying to recognize those changes.
Juan A. García-Pardo, J. Soler, C. Carrascosa

COLYPAN: A Peer-to-Peer Architecture for a Project Management Collaborative Learning System

In this paper, we present a project management collaborative learning system that tries to respond to the requirements of a motivating learning process. In this system, learner, group learners and tutors are in an environment where each one teaches and learns, by interacting with others. Peer-to-peer (p2p) network reflects and supports this relationship between users in a collaborative learning community. We propose a p2p agent-based system for their management and sharing.
Hanaa Mazyad, Insaf Tnazefti-Kerkeni

Preference Generation for Autonomous Agents

An intelligent agent situated in an environment needs to know the preferred states it is expected to achieve or maintain so that it can work towards achieving or maintaining them. We refer to all these preferred states as “preferences”. The preferences an agent has selected to bring about at a given time are called “goals”. This selection of preferences as goals is generally referred to as “goal generation”. Basic aim behind goal generation is to provide the agent with a way of getting new goals. Although goal generation results in an increase in the agent’s knowledge about its goals, the overall autonomy of the agent does not increase as its goals are derived from its preferences (which are programmed). We argue that to achieve greater autonomy, an agent must be able to generate new preferences. In this paper we discuss how an agent can generate new preferences based on analogy between new objects and the objects it has known preferences for.
Umair Rafique, Shell Ying Huang

Evaluation of Techniques for a Learning-Driven Modeling Methodology in Multiagent Simulation

There have been a number of suggestions for methodologies supporting the development of multiagent simulation models. In this contribution we are introducing a learning-driven methodology that exploits learning techniques for generating suggestions for agent behavior models based on a given environmental model. The output must be human-interpretable. We compare different candidates for learning techniques – classifier systems, neural networks and reinforcement learning – concerning their appropriateness for such a modeling methodology.
Robert Junges, Franziska Klügl

Applications II

Price Prediction in Sports Betting Markets

The sports betting market has emerged as one of the most lucrative markets in recent years. In this kind of prediction market, participants trade assets related to sports events according to their expectations. Prices in sports betting markets continually change depending on what is happening in the event. In this paper we propose an approach focused on predicting price movements in order to make benefits regardless of the final result.
We develop an agent who participates in the market focused on the task of learning the price movements in order to make predictions of future prices. Our approach is based on identifying and learn pattern price movements in order to predict the price movements of new events by using an underlying Case Based Reasoning system.
Juan M. Alberola, Ana Garcia-Fornes, Agustin Espinosa

Modelling Distributed Network Security in a Petri Net- and Agent-Based Approach

Distributed network security is an important concern in modern business environments. Access to critical information and areas has to be limited to authorised users. The Herold research project aims to provide a novel way of managing distributed network security through the means of agent-based software. In this paper we present the first models, both conceptual and technical that have been produced in this project. Furthermore we examine the Paose development approach used within the project and how it contributes to Herold.
Simon Adameit, Tobias Betz, Lawrence Cabac, Florian Hars, Marcin Hewelt, Michael Köhler-Bußmeier, Daniel Moldt, Dimitri Popov, José Quenum, Axel Theilmann, Thomas Wagner, Timo Warns, Lars Wüstenberg


Weitere Informationen

Premium Partner