Artifacts in the A&;A meta-model for multi-agent systems

In this article we focus on the notion of artifact for agents in multi-agent systems (MAS) as a basis for a new meta-model promoting the modelling and engineering of agent societies and MAS environment as first-class entities. Its conceptual foundations lay upon theories and results coming from computational sciences as well as from organisational and cognitive sciences, psychology, computer supported cooperative work (CSCW), anthropology and ethology. In the resulting agents & artifacts (A&A) meta-model, agents are the (pro-)active entities in charge of the goals/tasks that altogether build up the whole MAS behaviour, whereas artifacts are the reactive entities providing the services and functions that make individual agents work together in a MAS, and that shape agent environment according to the MAS needs. After presenting the scientific background, we define the notions of artifact in the A&A meta-model, discuss how it affects the notion of intelligence in MAS, and show its application to a number of agent-related research fields.     

AGENT COORDINATION CONTEXTS IN A MAS COORDINATION INFRASTRUCTURE

In order to harness complexity in multi-agent systems (MAS), first-class entities that mediate interaction between agents and environment are required, which can encapsulate control over MAS behavior and evolution. To this end, MAS infrastructures should provide mediating artifacts, both enabling and constraining agent interactions, and possibly representing admissible agent perceptions and actions over the environment.

Along this line, in this paper, we take the notion of agent coordination context (ACC) as a means to model agent-environment interactions, and show how it can be embedded within a MAS infrastructure in terms of model and runtime structures. Then, we take the TuCSoN coordination infrastructure as a reference, and extend it with the ACC abstraction to integrate the support for coordination with organization and security.     

A retraining methodology for enhancing agent intelligence

Data mining has proven a successful gateway for discovering useful knowledge and for enhancing business intelligence in a range of application fields. Incorporating this knowledge into already deployed applications, though, is highly impractical, since it requires reconfigurable software architectures, as well as human expert consulting. In an attempt to overcome this deficiency, we have developed Agent Academy, an integrated development framework that supports both design and control of multi-agent systems (MAS), as well as “agent training”. We define agent training as the automated incorporation of logic structures generated through data mining into the agents of the system. The increased flexibility and cooperation primitives of MAS, augmented with the training and retraining capabilities of Agent Academy, provide a powerful means for the dynamic exploitation of data mining extracted knowledge. In this paper, we present the methodology and tools for agent retraining. Through experimented results with the Agent Academy platform, we demonstrate how the extracted knowledge can be formulated and how retraining can lead to the improvement – in the long run – of agent intelligence.     

基于多Agent混合智能实现个性化网络信息推荐

1 引言互联网的迅速发展和普及,使得人们可以通过网络获取大量的信息资源,为帮助用户从互联网获取需要的各种信息资源,各种搜索引擎,如Yahoo!、Excite等相继诞生,它们为用户提供信息导航服务,帮助用户获取需要的信息。但随着网络资源的指数膨胀,使得搜索引擎难以为用户提供满意的服务。在这日益增多的     

A review of Multi-Agent Systems techniques, with application to Columbus User Support Organisation

This paper reviews an advanced Knowledge-Based Systems architecture based on multiple agents. An agent is a software entity with autonomous processing capabilities and a private database, which acts on its environment on the basis of information it receives from its environment, perceives, processes, retains and recalls. Multi-Agent Systems (MAS) are systems of agents which coordinate their knowledge, goals, skills and plans jointly to take action or to solve problems, including the problem of inter-agent coordination itself. This paper shows how such an architecture is being applied by a Dutch consortium to a part of the Columbus User Support Organisation (USO).

In the first part of the paper, Multi-Agent Systems techniques are placed into the context of Distributed Artificial Intelligence (DAI). Technical issues at the system architectural and agent structural levels are highlighted. The needs for intentionality (i.e. the ability of one agent to model another), acting, planning (both reactive and generative), and learning are identified.

The second part of the paper shows how MAS techniques may be applied to the In Orbit Infrastructure Ground Segment (IOI-GS) and, in particular, to the Columbus USO. The relevant features of the IOI-GS and the USO, including mission and experiment control hierarchies, are outlined, identifying potential applications of MAS techniques. The functions of a national User Support Operations Centre (USOC) are listed, and two functions are described in more detail, as they are expected to be implemented in The Netherlands' USOC (named the ‘Dutch Utilisation Centre’). A uniform MAS-based architecture is presented that is designed to support both functions. Prototype implementations of key parts of this architecture are described. Conclusions are drawn from the work performed so far.     

基于MAS Builder开发多Agent系统的方法与实例

随着计算机领域面临的应用问题越来越呈现出分布、开放、动态的特征，多Agent成为近年来AI研究的热点之一。本文简要介绍开发和集成多Agent系统的环境－MAS Builder，并以实例说明了基于该系统开发多Agent系统的方法：问题域分解、Agent定义、任务模块的生成。     

Limits of economic and strategic rationality for agents and MA systems

AI adoption of the game-theoretic paradigm although motivated and productive, suffers from basic limits for modelling autonomous agents and MA systems. First, we briefly restate game-theory's role for DAI and MAS: the introduction of formal prototypical social situations (“games”); the use of formal and sound notions, a self-interested view of autonomous agents, etc. Then, a number of criticisms, that have an impact on modelling intelligent social/individual action, are examined: the economicist interpretation of rationality; its instrumentalist conception, which leaves implicit the ends of agents' choices; the consequent multiple equilibria allowed by the theory; the context-unboundedness of rationality. Some contributions for a more heterarchic, context-bounded, architecture of rational agent are given, and a goal-based strategy, as distinct from a strictly utilitarian principle of decision-making, is proposed. Troubles of game-theory with multi-agent systems and in particular with modelling “cooperation” are outlined. Finally, some limits inherent in the notion of “incentive engineering” are pointed out.     

Refining imprecise data by integrity constraints

Uncertain data in databases were originally denoted as null values, which represent the meaning of ‘values unknown at present.” Null values were generalized into partial values, which correspond to a set of possible values, to provide a more powerful notion. In this paper, we derive some properties to refine partial values into more informative ones. In some cases, they can even be refined into definite values. Such a refinement is possible when there exist range constraint on attribute domains, or referential integrities, functional dependencies, or multivalued dependencies among attributes.

Our work actually eliminates redundant elements in a partial value. By this process, we not only provide a more concise and informative answer to users, but also speedup the computation of queries issued afterward. Besides, it reduces the communication cost when imprecise data are requested to be transmitted from one site to another site in a distributed environment.     

增强Q学习在非确定马尔可夫系统寻优问题中的应用

增强学习属于机器学习的一种,它通过与环境的交互获得策略的改进,其在线学习和自适应学习的特点使其成为解决策略寻优问题有力的工具。多智能体系统是人工智能领域的一个研究热点,对于多智能体学习技术的研究需要建立在系统环境模型的基础之上,由于多个智能体的存在,智能体之间的相互影响使得多智能体系统高度复杂,多智能体系统环境属于非确定马尔可夫模型,因此直接把基于马尔可夫模型的增强学习技术引入多智能体系统是不合适的。论文基于智能体间独立的学习机制,提出了一种改进的多智能体Q学习算法,使其适用于非确定马尔可夫环境,并对该学习技术在多智能体系统RoboCup中的应用进行了研究,实验证明了该学习技术的有效性与泛化能力,最后简要给出了多智能体增强学习研究的方向及进一步的工作。     

Solving weak transduction with EM

In this paper we describe an algorithm designed for learning perceptual organization of an autonomous agent. The learning algorithm performs incremental clustering of a perceptual input under reward. The distribution of the input samples is modeled by a Gaussian mixture density, which serves as a state space for the policy learning algorithm. The agent learns to select actions in response to the presented stimuli simultaneously with estimating the parameters of the input mixture density. The feedback from the environment is given to the agent in the form of a scalar value, or a reward, which represents the utility of a particular clustering configuration for the action selection. The setting of the learning task makes it impossible to use supervised or partially supervised techniques to estimate the parameters of the input density. The paper introduces the notion of weak transduction and shows a solution to it using an EM-based framework.     

Derivative-free reinforcement learning: a review

Reinforcement learning is about learning agent models that make the best sequential decisions in unknown environments. In an unknown environment, the agent needs to explore the environment while exploiting the collected information, which usually forms a sophisticated problem to solve. Derivative-free optimization, meanwhile, is capable of solving sophisticated problems. It commonly uses a sampling-andupdating framework to iteratively improve the solution, where exploration and exploitation are also needed to be well balanced. Therefore, derivative-free optimization deals with a similar core issue as reinforcement learning, and has been introduced in reinforcement learning approaches, under the names of learning classifier systems and neuroevolution/evolutionary reinforcement learning. Although such methods have been developed for decades, recently, derivative-free reinforcement learning exhibits attracting increasing attention. However, recent survey on this topic is still lacking. In this article, we summarize methods of derivative-free reinforcement learning to date, and organize the methods in aspects including parameter updating, model selection, exploration, and parallel/distributed methods. Moreover, we discuss some current limitations and possible future directions, hoping that this article could bring more attentions to this topic and serve as a catalyst for developing novel and efficient approaches.     

MAS体系结构在遥操作系统中的应用研究

设计了一个基于多智能体系统MAS结构的遥操作系统框架模型，它结合离散事件状态DES控制模型，可用于在目前广泛采用的将虚拟现实与自主智能系统相结合来克服时延影响的遥操作系统中，解决现场环境的几何学、动力学模型参数未知或不准确引起的相关问题。并以在遥操作系统中的直升飞机作为执行端为例，说明该系统框架模型的具体应用和可用性。     

群智能算法应用于MAS系统协作的探讨

多智能体系统(MAS)理论是目前人工智能领域的热点问题之一,群智能算法是一种并行式问题求解方法.就如何将群智能方法引入MAS研究中进行了探讨.首先对MAS理论的研究现状和发展趋势进行调研,综合考虑MAS中需要解决的问题和群智能算法的优点,认为两者具有结合应用的可行性.然后具体针对MAS协作方法研究中的通信瓶颈、意图解释机制、冲突消解等几类问题进行重点讨论和分析,就如何应用群智能算法进行了探讨,提出了初步的解决要点.     

Coordination Tools for MAS Development and Deployment

One of the most critical issues in the engineering of multi-agent systems (MAS) is the inadequacy of the available tools for MAS development and deployment. As we assume interaction as a first-class issue in MAS, tools are particularly required to monitor and debug inter-agents aspects, such as interaction protocols, coordination policies, social norms, and environment constraints. Since we claim that the definition of such tools is a basic research issue, in this paper we aim to identify the main requirements for development and deployment tools within an effective agent infrastructure. Focusing on agent interaction aspects, we take tuple-based coordination infrastructures--in particular the TuCSoN technology and tools--as our reference, and discuss the role of tools in a simple case study: the development and deployment of a well-known agent interaction protocol, the Contract Net.     

Competing agents in agent-mediated institutions

Social processes and agent interaction always take place in a specific context. A school of thought in social studies analyses them in the framework of institutions. We present in this paper the notion ofagentmediated institutions and show how it is relevant for multi-agent systems (MAS) in general and, more specifically, for MAS that include human agents and software agents involved in socioeconomic interactions. We show how the social interactions of human and software agents taking place in the Cohabited Mixed-Reality Information Spaces (COMRIS) project can be described as such an institution, the Conference Centre institution.     

On goal-directed provability in classical logic

One of the key features of logic programming is the notion of goal-directed provability. In intuitionistic logic, the notion of uniform proof has been used as a proof-theoretic characterization of this property. Whilst the connections between intuitionistic logic and computation are well known, there is no reason per se why a similar notion cannot be given in classical logic. In this paper we show that there are two notions of goal-directed proof in classical logic, both of which are suitably weaker than that for intuitionistic logic. We show the completeness of this class of proofs for certain fragments, which thus form logic programming languages. As there are more possible variations on the notion of goal-directed provability in classical logic, there is a greater diversity of classical logic programming languages than intuitionistic ones. In particular, we show how logic programs may contain disjunctions in this setting. This provides a proof-theoretic basis for disjunctive logic programs, as well as characterising the “disjunctive” nature of answer substitutions for such programs in terms of the provability properties of the classical connectives Λ and Λ.     

基于MAS(Multi Agent System)的多机器人系统：协作多机器人学发展的一个重要方向

机器人的应用方式正在由部件式单元应用向系统式应用方向发展．这是实际应用的需 要,也是技术发展的必然趋势;相关技术如计算机网络技术的发展也为它的实现提供了相应 支持．多机器人协作理论问题必然也已经成为机器人学研究的一个热点,其中,分布式人工 智能（DAI）中的多智能体（代理）系统（MAS：Multi-agent System）理论已引起多机器 人 协作理论研究者的关注．本文即在揭示协作多机器人系统与MAS的内在联系的基础上,指出 基于MAS的协作多机器人系统是协作多机器人学发展的一个重要方向．     

一种新的多智能体Q学习算法

针对非确定马尔可夫环境下的多智能体系统,提出了一种新的多智能体Q学习算法.算法中通过对联合动作的统计来学习其它智能体的行为策略,并利用智能体策略向量的全概率分布保证了对联合最优动作的选择. 同时对算法的收敛性和学习性能进行了分析.该算法在多智能体系统RoboCup中的应用进一步表明了算法的有效性与泛化能力.