Catálogo de publicaciones - libros

The performance of retrieving an image in terms of text-type of queries depends heavily on the quality of the annotated descriptive metadata that describes the content of the images. However, the effective annotation of an image can often be a laborious task that requires consistent domain knowledge. Annotators may annotate features in the images that could not contribute much to retrieval of the images. For effective annotation, an annotation guide agent (AGA) is proposed to aid annotators. Basically AGA monitors the annotator’s behaviors and based on the common sense induced from previous annotation instances as well as the domain ontology suggests critical property that will yield the most valuable information for image retrieval. We showed by experiments that the critical property and common sense heuristics used by AGA to aid the annotation of images could significantly lead to the improvement of the recall and precision of image retrieval.

Much current research is focussed on developing agent interaction protocols (AIPs) that will ensure seamless interaction amongst agents in multi agent systems. The research covers areas such as desired properties of AIPs, reasoning about interaction types, languages and tools for representing AIPs, and implementing AIPs. However, there has been little work on defining the structural make up of an agent interaction protocol, or defining dedicated approaches for developing agent interaction protocols from a clear problem definition to the final specification. This paper addresses these gaps. We present a dedicated approach for developing agent interaction protocols. Our approach is driven by an analysis of the application domain and our proposed structured agent interaction protocol definition.

This paper introduces teams of personal agents that support users individually in electronic negotiations. These agents listen to the running negotiation and to each other to point out relevant information and compile advice for the user. In this frame, we first describe the architecture of this system and propose assistance interaction protocols to specify agent external behaviours in performing their tasks. Then, we discuss the semantic representation of agent communication and describe an abstraction layer to let agents understand user message issues. Our future work aims at improving these mechanisms and enriching them toward a full-fledged implementation.

This paper describes the architecture of a distributed workflow management system in a dynamic environment. The system features autonomous agent components that can adapt to both structural changes in business processes and changes in system parameters, such as the number of available resources. This adaptation could be a permanent adjustment that should be reflected in all the incoming work cases, or be associated with a particular instance of a work case. In addition, parts of the system can be modified by observing the behaviour of the system for possible shortcomings due to a non-optimal distribution of resources or faulty inter-process dependencies which could result in bottlenecks. Because of the autonomous nature of subsystem components, the workflow system can adapt to changes without the necessity of centralized control. The architecture of the system is described in the context of a distributed workflow example. Keywords: dynamic workflow, autonomous components, interaction protocols, coloured Petri nets, adaptability

Workflow management systems (WfMS) should address the needs of rapidly changing business environments. We have built a multi-agent based framework, JBees, which addresses these needs. We evaluate our agent-based workflow system, which employs coloured Petri net workflow modeling, with the proposed standards for various workflow patterns and communication patterns. The coloured Petri net models support the workflow patterns and the agent-based framework supports the communication standards developed by the Foundation for Intelligent Physical Agents (FIPA). The agent-based communication technology patterns along with the workflow patterns equip the workflow management system with a comprehensive set of capabilities, such as adaptability and distribution.

Business and engineering systems are increasingly being created as collections of many autonomous (human or software) agents cooperating as peers. Peer-to-peer coordination introduces, however, unique and potentially serious challenges. When there is no one ‘in charge’, dysfunctions can emerge as the collective effect of locally reasonable decisions. In this paper, we consider the dysfunction wherein inefficient resource use oscillations occur due to delayed status information, and describe novel approaches, based on the selective use of misinformation, for dealing with this problem.

We are concerned with forms of interaction in which multiple users, with differing agendas and interests, may realise opportunities for useful synchronisation of their activities. We present a framework in which intelligent software agents act as semi-autonomous intermediaries among nomadic users. Agents capture and process information about situations (specifically about the environment, users and their activities) in order to jointly find and negotiate opportunities for coordinating the activities of their respective users. The interaction is structured using a negotiation protocol that exploits a hierarchical representation of tasks and goals.

This paper proposes a formal approach, based on stepwise refinements, for specifying and designing multi-agent applications. This approach provides a specification language which integrates temporal logic in the Z notation allowing, in this way, to cover static, behavioural, as well as dynamic aspects of multi-agent systems. Moreover, it proposes a methodology giving a set of hints and principles which help and guide the design process. Indeed, this methodology enables the user to develop step by step, in an incremental way, an implementation starting from an abstract requirements (goal) specification. Finally, we illustrate our approach by developing an agent based solution for the pursuit problem .

This paper presents an approach to developing and managing self-organizing distributed computing systems. The approach is used to construct an application as a dynamic federation of mobile components that can migrate from computer to computer while the application is being executed. It also enables each component to explicitly define its own migration policy as the migration of other components. Therefore, a federation of components can be migrated and transformed according to its components’ local policies, including bio-inspired deployment approaches. The approach was implemented as not only a test-bed system for the organization of multi-agents but also a middleware for real distributed systems. This paper describes a prototype implementation of the middleware built on a Java-based mobile agent system and its applications that illustrates the utility and effectiveness of the approach.

It is a common behavior that a group of rational agents cooperate together as a bidder/seller to bid in an auction. How to determine the group bidding price and how to share the profit among the members in a group has been problems that are not studied thoroughly. In time-critical auctions, the problem is getting more complicated since the group has to decide new bidding prices within time limits. Conventional approaches used a centralized mechanism to assign profit share to each bidding agent in the group that usually lead to negative profit of individual bidding agent. We propose a distributed approach called Z-process that allows individual bidding agents to declare their compromised profit share based on their rationalities, and determines the group bidding prices simultaneously. We show that in Z-process there exists a dominant strategy for rational agents that can let them obtain maximum profit. We can also show that the compromised profit of each individual bidding agent by Z-process satisfies each agent’s rationality.