Innovations in Multi-Agent Systems and Applications - 1

Multi-agent systems is a subfield of Distributed Artificial Intelligence that has experienced rapid growth because of the flexibility and the intelligence available solve distributed problems. In this chapter, a brief survey of multi-agent systems has been presented. These encompass different attributes such as architecture, communication, coordination strategies, decision making and learning abilities. The goal of this chapter is to provide a quick reference to assist in the design of multi-agent systems and to highlight the merit and demerits of the existing methods.

Hybrid systems have grown tremendously in the past few years due to their abilities to offset the demerits of one technique by the merits of another. This chapter presents a number of computational intelligence techniques which are useful in the implementation of hybrid multi-agent systems. A brief review of the applications of the hybrid multi-agent systems is presented.

In this chapter, the Coordinated Hybrid Agent (CHA) framework is introduced for the distributed control and coordination of multi-agent systems. In this framework, the control of multi-agent systems is regarded as achieving decentralized control and coordination of agents. Each agent is modeled as a CHA which is composed of an intelligent coordination layer and a hybrid control layer. The intelligent coordination layer takes the coordination input, plant input and workspace input. The intelligent coordination layer deals with the planning, coordination, decision-making and computation of the agent. The hybrid control layer of the framework takes the output of the intelligent coordination layer and generates discrete and continuous control signals to control the overall process. In order to verify the feasibility of the framework, experiments for multi-agent systems are implemented. The framework is applied to a multi-agent system consisting of an overhead crane, a mobile robot and a robot manipulator. The agents are able to cooperate and coordinate to achieve the global goal. In addition, the stability of systems modeled using the framework is also analyzed.

During the summer (2000), wholesale electricity prices in California were approximately 500% higher than those during the same months in 1998-1999. This study proposes a practical use of a reengineered Multi-Agent Intelligent Simulator (MAIS) to numerically examine several reasons on why the crisis has occurred during May 2000-Janurary 2001. The proposed MAIS generates artificially numerous trading agents equipped with different learning capabilities and duplicates their bidding strategies in the California electricity markets during the crisis period. In this study, we confirm the methodological validity of MAIS by comparing the estimation accuracy of MAIS with those of the three well-known computer science techniques (Support Vector Machines, Neural Networks and Genetic Algorithms). This study also investigates the dynamic change on agent composition in a time horizon. This investigation finds that all agents gradually shift to multiple learning capabilities so as to adjust themselves to the price fluctuation of electricity. Finally, we apply the sensitivity analysis of MAIS to identify economic rationales concerning the crisis. The sensitivity analysis results in the estimation accuracy (91.15%) during the crisis period. This study finds that 40.46% of the price increase during the crisis period was due to an increase in marginal production cost, 17.85% to traders’ greediness, 5.27% to a real demand change and 3.56% to market power. The remaining 32.86% came from other unknown market fundamentals and an estimation error. This numerical result indicates that the price hike has occurred due to an increase in fuel prices and real demand. The change of the two market fundamentals explained 45.73% (= 40.46% + 5.27%) of the price increase and fluctuation during the crisis. The responsibility of energy utility firms was 21.41% (= 17.85% + 3.56%).

Argumentation is an interdisciplinary research area that incorporates many fields such as artificial intelligence, multi-agent systems, and collaborative learning. In this chapter, we describe argument mining techniques from a structured argument database “RADB”, a sort of relational database we designed specially for organizing argument databases, and their usage in arguing agents and intelligent tutoring systems. The RADB repository depends on the Argumentation Interchange Format Ontology (AIF) using “Walton Theory” for argument analysis. It presents a novel approach that summarizes the argument data set into structured form “RADB” in order to (i) facilitate the data interoperability among various agents/humans/tools, (ii) provide the ability to freely navigate the repository by integrating the data mining techniques gathered in a classifier agent; mine the RADB repository and retrieve the most relevant arguments to the users’ queries, (iii) illustrate an agent-based learning environment outline, where the mining classifier agent and the RADB are incorporated together within an intelligent tutoring system (ITS). Such incorporation assists in (i) deepening the understanding of negotiation, decision making, and critical thinking, (ii) guiding the analysis process to refine the user’s underlying classification, and improving the analysis and the students’ intellectual process.

Later in the chapter, we describe an effective usage of argument mining for arguing agents, which interact with each other in the Internet environment and argues about issues concerned, casting arguments and counter-arguments each other to reach an agreement. We illustrate how argument mining allows to strengthen arguing agent intelligence, resulting in expanding the main concern in formal argumentation frameworks that is to formalize methods in which the final statuses of arguments are to be decided semantically and/or dialectically. In both usages, we yield new forms of argument-based intelligence, which allows establishing one’s own argument by comparing diverse views and opinions and uncovering new leads, differently from simple refutation aiming at cutting down other parties.

Several models have been proposed for describing grouping behavior such as bird flocking, terrestrial animal herding, and fish schooling. In these models, a fixed rule has been imposed on each individual a priori for its interactions in a reductive and rigid manner. We have proposed a new framework for self-organized grouping of agents by reinforcement learning. It is important to introduce a learning scheme for developing collective behavior in artificial autonomous distributed systems. This scheme can be expanded to cases in which predators are present. In this study we integrate grouping and anti-predator behaviors into our proposed scheme. The behavior of agents is demonstrated and evaluated in detail through computer simulations, and their grouping and anti-predator behaviors developed as a result of learning are shown to be diverse and robust by changing some parameters of the scheme.

Multi-agent systems can be used to address problems in a variety of domains, including robotics, distributed control, telecommunications, and economics. The complexity of many tasks arising in these domains makes them difficult to solve with preprogrammed agent behaviors. The agents must instead discover a solution on their own, using learning. A significant part of the research on multi-agent learning concerns reinforcement learning techniques. This chapter reviews a representative selection of multi-agent reinforcement learning algorithms for fully cooperative, fully competitive, and more general (neither cooperative nor competitive) tasks. The benefits and challenges of multi-agent reinforcement learning are described. A central challenge in the field is the formal statement of a multi-agent learning goal; this chapter reviews the learning goals proposed in the literature. The problem domains where multi-agent reinforcement learning techniques have been applied are briefly discussed. Several multi-agent reinforcement learning algorithms are applied to an illustrative example involving the coordinated transportation of an object by two cooperative robots. In an outlook for the multi-agent reinforcement learning field, a set of important open issues are identified, and promising research directions to address these issues are outlined.

One of the main characteristics of multi-agent systems (MAS) is fault tolerance. When an agent is unavailable for some reason, another agent with similar capabilities can theoretically compensate for this loss. Many key aspects of fault tolerance in MAS are described in this chapter including social knowledge, physical distribution, agent development, and validation. Therefore, the focus is not only on a fault tolerant agent platform with necessary services (e.g., fault tolerant social knowledge), but also on the design that can significantly reduce mistakes in agent programming and validation that can discover faults that manifest as failures during the testing phase.

In recent decades, agent-based modeling and simulation (ABMS) has been increasingly used as a valuable approach for design and analysis of dynamic and emergent phenomena of large-scale, complex multi-agent systems, including socio-technical systems. The dynamic behavior of such systems includes both the individual behavior of heterogeneous agents within the system and the emergent behavior arising from interactions between agents within their work environment; both must be accurately modeled and efficiently executed in simulations. An important issue in ABMS of socio-technical systems is ensuring that agents are updated together at any time where they must interact or exchange data, even when the agents’ internal models use fundamentally different methods of advancing their internal time and widely varying update rates. This requires accurate predictions of interaction times between agents within the environment. Predicting the time of interactions, however, is not a trivial problem. Thus, timing mechanisms that advance simulation time and select the proper agent to be executed are crucial to correct simulation results. This chapter describes a timing and prediction mechanism for accurate modeling of interactions among agents which also increases the computational efficiency of agent-based simulations. An experiment comparing different timing methods highlighted the gains in computational efficiency achieved with the new timing mechanisms and also emphasized the importance of identifying correct interaction times. An intelligent timing agent framework for predicting the timing of interactions between heterogeneous agents using a neural network and a method for assessing the accuracy of interaction prediction methods based on signal detection theory are described. An application of agent-based modeling and simulation to air transportation systems serves as a test case and the simulation results of different interaction prediction models are presented. The insights of using the framework and method to the design and analysis of complex socio-technical systems are discussed.

The chapter deals with group-oriented service provisioning in next-generation network (NGN). It consists of three parts: the first bringing forth user profile creation and semantic comparison; the second explaining user profile clustering and semantic classification; and the third describing social network creation and analysis. The multi-agent system A-STORM (Agent-based Service and Telecom Operations Management) is presented and elaborated as part of the proof-of-concept prototype which demonstrates provisioning of group-oriented services within NGN. As a group-oriented service, the RESPIRIS (Recommendation-based Superdistribution of Digital Goods within Implicit Social Networks) service is implemented and provisioned by using prototype’s agents. The proposed provisioning scenario is set forth, as well as provisioning process analysis presented.