Catálogo de publicaciones - libros

In this paper, we describe our agent framework and address the issues we have encountered designing a suitable environmental space for evaluating the coordination and adaptive qualities of multi-agent systems. Our research direction is to develop a framework allowing us to build different type of agents rapidly, and to facilitate the addition of new technology. The underlying technology of our Java Agent Framework (JAF) uses a component-based design. We will present in this paper, the reasons and the design choices we made to build a complete system to evaluate the coordination and adaptive qualities of multi-agent systems.

Design-to-Criteria builds custom schedules for agents that meet hard temporal constraints, hard resource constraints, and soft constraints stemming from soft task interactions or soft commitments made with other agents. Design-to-Criteria is designed specifically for online application - it copes with exponential combinatorics to produce these custom schedules in a resource bounded fashion. This enables agents to respond to changes in problem solving or the environment as they arise.

In this paper we identify several types of specific services that agent infrastructures must support, including complex interaction, decision making for individual utility maximization, team formation in organizations and multi-attribute negotiation. We describe each of these services in part and show how they are supported in an integrated manner to provide abstract functionality related to complex interaction and unified individual and social reasoning in multi-agent settings. We show that this range of services can be achieved by combining conversational technology at the interaction level with constraint satisfaction and optimization as a common reasoning infrastructure supporting the required reasoning services. All services are provided by generic components packaged into a Java written Agent Building Shell.

This paper presents a tool that enables the bottom-up design of multi-agent systems. The tool has two parts. The first part is a wrapper that wraps each agent so that it exempts the designers from the careful detailed deployment of the inter-relationships between cooperation knowledge and task knowledge inside the agent. This wrapper should be independent of the functions of agents. The second part is an environment that can support the wrapper to automate the cooperation process in behalf of agents.

This paper describes an experiment with mobility in multi-agent systems. The setting is a virtual machine that supports reification of the computational state of a running process. The objective is to investigate how this feature facilitates telescripting and to speculate on how languages like Java should evolve to include the resulting notion of strong migration

A major challenge for distributed applications working in mobile contexts is to provide application developers with a method of building stable systems whose elements may change across time. We introduce the concept of As Strong As Possible mobility that uses a combination of data space management and thread state capture so that objects and threads can migrate in a manner that has not been properly explored yet. The ultimate goal is to provide a mechanism for mobility where an object will be migrated using strong mobility techniques where possible and using rebinding mechanisms when it is not advantageous to simply ’grab’ a thread’s state.

In the last three years, we have developed SETA, a prototype toolkit for the creation of Web stores personalizing the interactions with users, focusing on the design of the front-end of on-line stores and on the development of a flexible interface. We have exploited knowledge representation techniques and agent-based technologies to improve the configurability of the toolkit and its scalability [,]. Moreover, we have organized the overall architecture as a multiagent one, where specialized agents fill the main roles for the management of personalized interactions with customers []. An on-line demo of a prototype store created using SETA is available at the following URL: http://www.di.unito.it/~ seta. This store presents telecommunication products, like phones and switchboards, and will be used throughout the rest of the paper as a concrete example to describe the functionalities of our system.

In this paper we propose a novel performance analysis approach that can be used to gauge quantitatively the performance characteristics of different mobile-agent platforms. We materialize this approach as a hierarchical framework of benchmarks designed to isolate performance properties of interest, at different levels of detail. We identify the structure and parameters of benchmarks and propose metrics that can be used to capture their properties. We present a set of micro-benchmarks, comprising the lower level of our hierarchy, and examine their behavior when implemented with commercial, Java-based, mobile agent platforms.

The development of agent systems based on layered agent templates has expanded the development and deployment of agent applications. We are developing an agent template to support the integration of a wide range of enterprise applications. In particular, we are targeting existing enterprises, such that agents can manage enterprise resources and reason across these resources. This agent template has five functional layers. Software developers can plug-and-play layers and customize features in each layer. We have discovered infrastructure requirements and design issues for employing layered agent designs into existing enterprises. In this brief paper we describe some requirements of our agent template. In general, we believe these same issues are relevant to system designers who are building multi-agent systems for use in large scale enterprise computing.

A community of agent system is proposed for administration and control of remote instrumentation in the Materials Microcharacterization Collaboratory (MMC). The MMC is a joint-project between national laboratories, academia and industrial collaborators in the US. The community of agents makes use of software agents for user authorization and authentication, training, scheduling instrumentation, and responding to user queries. A methodology providing for agent responsibilities and an interaction model is being used. Scalability issues discussed include raising the number of instruments and of users in the system. The authorization and authentication agent creates a bottleneck.