José M. Vidal

Online auctions

Auctions on the Internet can involve not only consumers, but also businesses. They can form dynamically and enable the exchange of goods much as stock exchanges manage the buying and selling of securities. But because auctions have a wide scope and a short lifetime, the opportunistic behavior needed for successful interaction requires agents to both participate in and manage auctions. The article focuses on the use of software agents in such Internet based auctions.

Multiagent systems with workflows

Industry and researchers have two different visions for the future of Web services. Industry wants to capitalize on Web service technology to automate business processes via centralized workflow enactment. Researchers are interested in the dynamic composition of Web services. We show how these two visions are points in a continuum and discuss a possible path for bridging the gap between them.

Toward inquiry-based education through interacting software agents

The University of Michigan Digital Library (UMDL) project is creating an infrastructure for rendering library services over a digital network. When fully developed, the UMDL will provide a wealth of information sources and library services to students, researchers, and educators. Tasks are distributed among numerous specialized modules called agents. The three classes of agents are user interface agents, mediator agents, and collection interface agents. Complex tasks are accomplished by teams of specialized agents working together-for example, by interleaving various types of search. The UMDL is being deployed in three arenas: secondary-school science classrooms, the University of Michigan library, and space-science laboratories. The development team expects the scale and diversity of the project to test their technical ideas about distributed agents, interoperability, mediation, and economical resource allocation.

Supply network topology and robustness against disruptions - An investigation using multi-agent model

In this study we examine the relationship between supply networks topology and its robustness in the presence of random failures and targeted attacks. The agent-based model developed in this paper uses the basic framework and parameters in the experimental game presented in Sterman [1989, Modeling managerial behavior: Misperceptions of feedback in a dynamic decision making context. Management Science, 35 (3), 321–339] for modelling adaptive managerial decision making in an inventory management context. The study extends the linear supply chain context to a complex supply network and undertakes a rigorous examination of robustness of these supply networks that are characterised by distinct network characteristics. We theorise that network characteristics such as average path length, clustering coefficient, size of the largest connected component in the network and the maximum distance between nodes in the largest connected component are related to the robustness of supply networks, and test the research hypotheses using data from several simulation runs. Simulations were carried out using 20 distinct network topologies where 10 of these topologies were generated using preferential attachment approach (based on the theory of scale-free networks) and the remaining 10 topologies were generated using random attachment approach (using random graph theory as a foundation). These 20 supply networks were subjected to random demand and their performances were evaluated by considering varying probabilities of random failures of nodes and targeted attacks on nodes. We also consider the severity of these disruptions by considering the downtime of the affected nodes. Using the data collected from a series of simulation experiments, we test the research hypotheses by means of binomial logistic regression analysis. The results point towards a significant association between network characteristics and supply network robustness assessed using multiple performance measures. We discuss the implications of the study and present directions for future research.

Matchmaking of web services based on the DAML-S service model

DAML-S provides the means for a web service to advertise its functionality to potential users of the service. This brings to fore the issue of discovering an advertisement that best matches a request for a particular service---a process referred to as matchmaking. The algorithms that have thus far been proposed for matchmaking are based on comparisons of the requested and offered inputs and outputs. In this project, we extend these algorithms by taking into account the detailed process description of the service, thus leading to more accurate matchmaking.

Learning nested agent models in an information economy

Abstract. We present our approach to the problem of how an agent, within an economic multi-agent system, can determine when it should behave strategically (i.e. learn and use models of other agents), and when it should act as a simple price-taker. We provide a framework for the incremental implementation of modelling capabilities in agents, and a description of the forms of knowledge required. The agents were implemented and different populations simulated in order to learn more about their behaviour and the merits of using and learning agent models. Our results show, among other lessons, how savvy buyers can avoid being ‘cheated’ by sellers, how price volatility can be used to quantitatively predict the benefits of deeper models, and how specific types of agent populations influence system behaviour.

Inside an agent

When we discuss agent-based system construction with software developers or ask students to implement common agent architectures using object-oriented techniques, we find that it is not trivial for them to create an elegant system design from the standard presentation of these architectures in textbooks or research papers. To better communicate our interpretation of popular agent architectures, we draw UML (Unified Modeling Language) diagrams to guide an implementers design. However, before we describe these diagrams, we need to review some basic features of agents. The paper considers an architecture showing a simple agent interacting with an environment. The agent senses its environment, uses what it senses to choose an action, and then performs the action through its effectors. Sensory input can include received messages, and action can be the sending of messages. To construct an agent, we need a more detailed understanding of how it functions. In particular, if we are to build one using conventional object-oriented analysis and design techniques, we should know in what ways an agent is more than just a simple object.

Predicting the Expected Behavior of Agents that Learn About Agents: The CLRI Framework

We describe a framework and equations used to model and predict the behavior of multi-agent systems (MASs) with learning agents. A difference equation is used for calculating the progression of an agents error in its decision function, thereby telling us how the agent is expected to fare in the MAS. The equation relies on parameters which capture the agents learning abilities, such as its change rate, learning rate and retention rate, as well as relevant aspects of the MAS such as the impact that agents have on each other. We validate the framework with experimental results using reinforcement learning agents in a market system, as well as with other experimental results gathered from the AI literature. Finally, we use PAC-theory to show how to calculate bounds on the values of the learning parameters.

Learning in multiagent systems: an introduction from a game-theoretic perspective

We introduce the topic of learning in multiagent systems. We first provide a quick introduction to the field of game theory, focusing on the equilibrium concepts of iterated dominance, and Nash equilibrium. We show some of the most relevant findings in the theory of learning in games, including theorems on fictitious play, replicator dynamics, and evolutionary stable strategies. The CLRI theory and n-level learning agents are introduced as attempts to apply some of these findings to the problem of engineering multiagent systems with learning agents. Finally, we summarize some of the remaining challenges in the field of learning in multiagent systems.

The effects of co-operation on multiagent search in task-oriented domains

This paper studies the benefits of teaming and selflessness when using multiagent search to solve task-oriented problems. A formal framework for multiagent search is presented, which forms a superset of the task-oriented domain, coalition formation, distributed constraint satisfaction and NK landscape search problems. The paper focuses on task-oriented domain problems and shows how the benefits of teaming and selflessness arise in this domain. These experimental results are compared to similar results in the NK domain—from which a predictive technique is imported. Namely, it is shown that better allocations are found when the dynamics of the multiagent system lie between order and chaos. Several other specific findings are presented such as the fact that neither absolute selfishness nor absolute selflessness result in better allocations, and the fact that the formation of small teams usually leads to better allocations.