A learning method of immune multi-agent neural networks

Many different learning algorithms for neural networks have been developed, with advantages offered in terms of network structure, initial values of some parameters, learning speed, and learning accuracy. If we train the networks only on good examples, without noise and shortage, the neural network can be trained to classify, with reasonable accuracy, target patterns or random patterns, but not both. To solve this problem, we propose a learning method of immune multi-agent neural networks (IMANNs), which have agents of macrophages, B-cells and T-cells. Each agent employs a different type of neural network. Because the agents work cooperatively and competitively, IMANNs can automatically classify the training dataset into some subclasses. In this paper, two types of IMANNs are described and their classification capabilities are compared. In order to verify the effectiveness of our proposed method, we used two datasets: the dataset of the MONKs problem (as a traditional classification problem) and a dataset from a medical diagnosis problem (hepatobiliary disorders).     

Fast Online Q(λ)

Q()-learning uses TD()-methods to accelerate Q-learning. The update complexity of previous online Q() implementations based on lookup tables is bounded by the size of the state/action space. Our faster algorithm's update complexity is bounded by the number of actions. The method is based on the observation that Q-value updates may be postponed until they are needed.     

Agent evaluation based on multi-source heterogeneous information table using TOPSIS

In agent evaluation, a specific role-playing may need more than one capabilities or the task execution process can be divided into several stages. The diverse perspectives to assess candidate agents are denoted as attributes, which is more practical than treating experts as attributes in many other works. In the evaluation table, the attribute values may come from different sources and the data types may not be the same. Therefore, we consider evaluation issues in a Multi-Source Heterogeneous Information System (MSHIS). Considering that grading, voting and marking are three common evaluation scenarios, linguistic variable, Intuitionistic Fuzzy Value (IFV) and real number are utilized to describe the corresponding evaluation results. To evaluate agents in MSHIS, a TOPSIS-based evaluation method is adopted in this work. In the proposed method, the range is utilized to nondimensionalize the distance between agents in each attribute. Then, a weighted Euclidean distance metric is adopted to measure the comprehensive distance. The relative closeness to the ideal agents reflects the agent’s capability on the concerned task. Finally, the illustrative example and comparative experiments are presented to illustrate the effectiveness of our method.     

SANet: a service-agent network for call-center scheduling

We consider a network of service-providing agents, where different agents have different capabilities, availability, and cost to solve problems. These characteristics are particularly important in practice for semi-automated call centers which provide quality customer service in real time. We have developed SANet, a service agent network for call center automation, to serve as an experimental testbed for our research. SANet can select appropriate agents to provide better solutions for customer problems according to the changing capabilities and availability of service agents in the network. It can also add or delete appropriate agents to balance problem-solving quality, efficiency, and cost according to the number and types of incoming customer problems. On this network, each service agent can be a human service agent, an automated software service agent, or a combination of the two. This paper describes the architecture, a problem scheduling algorithm and an agent assignment algorithm on the SANet. We highlight an application in which we apply SANet to a call-center scheduling problem for a cable-TV company. Finally, we show the efficiency and adaptability of our system via experimental results and discuss related works.     

On calculation of the power indices with allowance for the agent preferences

In the general case, complexity of the algorithm to calculate the power indices grows exponentially with the number of voting agents. Yet the volume of calculations may be reduced dramatically if many coalitions have equal numbers of votes. The well-known algorithm for calculation of the Banzhaf and Shapley-Shubik indices was generalized, which enables fast calculation of the power indices where entry of the voting agent into a coalition depends on its preferences over the set of the rest of agents.     

On the Relationship Between BDI Logics and Standard Logics of Concurrency

The behavior of an agent is mainly governed by the specific way in which it handles the rational balance between information and deliberation. Rao and Georgeff's BDI theory is most popular among the formalisms capturing this very balance. This formalism has been proposed as a language for specifying agents in an abstract manner or, alternatively, for verifying various properties of agents implemented in some other programming language. In mainstream computer science, there are formalisms designed for a purpose similar to the BDI theory; not specifically aiming at agents, but at concurrency in general. These formalisms are known as logics of concurrent programs. In this paper these two frameworks are compared with each other for the first time. The result shows that the basic BDI theory, BDI_CTL*, can be captured within a standard logic of concurrency. The logic which is relevant here is Kozen's propositional -calculus. -calculus turns out to be even strictly stronger in expressive power than BDI_CTL* while enjoying a computational complexity which is not higher than that of BD_CTL*'s small fragment CTL. This correspondence puts us in a position to provide the first axiomatization of Rao and Georgeff's full theory. Immediate consequences for the computational complexity of BDI theory are also explored, both for theorem proving and model checking.     

合作agent的能力描述

能力描述是agent合作的基础，本文提出了一个通用的MAS模型，分析了合作agent的能力描述方案，给出了一个agent能力描述语言（ACDL），能够对个体agnet，一组agent以及agency的能力进行描述，支持不同规模的agent合作和基于能力的学习。     

ON THE ROLE OF BDI MODELING FOR INTEGRATED CONTROL AND COORDINATED BEHAVIOR IN AUTONOMOUS AGENTS

This paper describes an architecture for controlling and coordinating autonomous agents, building on previous work addressing reactive and deliberative control methods. The proposed multilayered hybrid architecture allows a rationally bounded, goal-directed agent to reason predictively about potential conflicts by constructing knowledge level models that explain other agents' observed behaviors and hypothesize their beliefs, desires, and intentions; at the same time, it enables the agent to operate autonomously, to react promptly to changes in its real-time environment, and to coordinate its actions effectively with other agents. A principal aim of this research is to understand the role dzfferent functional capabilities play in constraining an agent5 behavior under varying environmental conditions. To this end, an experimental test bed has been constructed comprising a simulated multi-agent world in which a variety of agent configurations and behaviors have been investigated. A number of experimentalfindings are reported.     

EVA: AN EVOLUTIONARY APPROACH TO MUTUAL MONITORING OF LEARNING INFORMATION AGENTS

A learning agent system is composed of agents able to autonomously enrich their knowledge and improve their performance, using learning strategies. The idea underlying this article is that individual improvements obtained by the learning capabilities of an agent should be exploited to advantage the other agents, and a natural way of obtaining such a result is represented by evolutionary processes. However, the biological evolutionary mechanisms are often too complex to be reproduced in a software environment. In this context, we argue that cloning, due to its very simple mechanism of reproduction, can be usefully used. In our approach, a user in a virtual community can substitute an unsatisfactory agent cloning an existing agent having both similar interests and a good reputation in the community. This mechanism induces an evolutionary process in the community, such that the less satisfactory agents are replaced by more effective agents. The key issue of this proposal is that of suitably selecting the agent to be cloned in the presence of a user's request, and to this purpose we propose an evolutionary model of reputation. Our evolutionary approach has been implemented on top of a leaning agent-based recommender system, and a number of experiments show that this novel strategy introduces significant improvements in the effectiveness of the recommendations.     

移动Agent可靠位置透明通信方法的研究

在Mobile Agent 的应用中,Agent不是孤立地完成任务,而是需要不断地与其它Agent进行协作和信息交换。而Mobile Agent的特性使它的位置又经常变动。文章论述了一种实现Mobile Agent位置透明通信的方法,主要解决Agent位置追踪问题和Agent迁移时的消息处理。     

Studies on a surface acoustic wave (SAW) dosimeter sensor for organophosphorous nerve agents

As a follow-up of previous work on a Surface Acoustic Wave (SAW) sensor for nerve agents, irreversible response effects have been studied in more detail. Surface analytical studies indicated that degradation products are responsible for the effects observed. In addition it was tried to explore these effects for the development of a nerve agent dosimeter. Experiments were conducted to test the performance of a SAW sensor coated with La(III) 2-bis(carboxymethyl)amino hexadecanoic acid. The experiments revealed that many improvements must be made especially with respect to sensitivity and linear response behaviour.     

离线优化在线查表的双层结构预测控制

双层结构预测控制包括稳态优化和动态控制,计算复杂度较大,难以在实时性要求较高或者是主控制器计算能力较弱的场合应用.本文阐述了一种离线计算在线查表的稳态优化方法和详细的技术实施方案.首先采用枚举方式离线计算稳态优化值,然后通过在线查表得到实时优化值.当在线查表不可行时,通过查找距离该查表点最近的多个可查表点,以距离作为加权系数,近似计算最优稳态目标值.最后,通过性能分析和仿真实验证明了本文提出的该方法有效的降低了计算复杂度,减少了计算时间.     

A computational model for distributed knowledge systems with learning mechanisms

This paper addresses the issues of machine learning in distributed knowledge systems, which will consist of distributed software agents with problem solving, communication and learning functions. To develop such systems, we must analyze the roles of problem-solving and communication capabilities among knowledge systems. To facilitate the analyses, we propose a computational model: LPC. The model consists of a set of agents with (a) a knowledge base for learned concepts, (b) a knowledge base for problem solving, (c) prolog-based inference mechanisms and (d) a set of beliefs on the reliability of the other agents. Each agent can improve its own problem-solving capabilities by deductive learning from the given problems, by memory-based learning from communications between the agents and by reinforcement learning from the reliability of communications between the other agents. An experimental system of the model has been implemented in Prolog language on a Window-based personal computer. Intensive experiments have been carried out to examine the feasibility of the machine learning mechanisms of agents for problem-solving and communication capabilities. The experimental results have shown that the multiagent system improves the performance of the whole system in problem solving, when each agent has a higher learning ability or when an agent with a very high ability for problem solving joins the organization to cooperate with the other agents in problem solving. These results suggest that the proposed model is useful in analyzing the learning mechanisms applicable to distributed knowledge systems.     

Cooperative reinforcement learning in topology-based multi-agent systems

Topology-based multi-agent systems (TMAS), wherein agents interact with one another according to their spatial relationship in a network, are well suited for problems with topological constraints. In a TMAS system, however, each agent may have a different state space, which can be rather large. Consequently, traditional approaches to multi-agent cooperative learning may not be able to scale up with the complexity of the network topology. In this paper, we propose a cooperative learning strategy, under which autonomous agents are assembled in a binary tree formation (BTF). By constraining the interaction between agents, we effectively unify the state space of individual agents and enable policy sharing across agents. Our complexity analysis indicates that multi-agent systems with the BTF have a much smaller state space and a higher level of flexibility, compared with the general form of n-ary (n > 2) tree formation. We have applied the proposed cooperative learning strategy to a class of reinforcement learning agents known as temporal difference-fusion architecture for learning and cognition (TD-FALCON). Comparative experiments based on a generic network routing problem, which is a typical TMAS domain, show that the TD-FALCON BTF teams outperform alternative methods, including TD-FALCON teams in single agent and n-ary tree formation, a Q-learning method based on the table lookup mechanism, as well as a classical linear programming algorithm. Our study further shows that TD-FALCON BTF can adapt and function well under various scales of network complexity and traffic volume in TMAS domains.     

一种基于多主体技术的分布式图像聚类算法

针对医学图像具有对比度较低,不同组织之间的模糊性较高的特点,给出一种基于多主体和数学形态学灰度形态运算的聚类算法。算法采用agent技术和多结构元素结合的模式,用结构元素做智能个体,每个不同类型的agents随机散布在离散空间格点上,在同时刻控制系统驱动下agents根据其自身结构元素的类型用给出的邻域平均算子自主选择作相应的运算进而实现图像聚类。算法无须先验知识和预处理操作,对初始聚类点不敏感,无须事先输入聚类簇数。算法具有分布式并行计算功能和自主分析能力。实验结果验证了该算法的可行性和可靠性。     

Reasoning about action and change

Reasoning about change is a central issue in research on human and robot planning. We study an approach to reasoning about action and change in a dynamic logic setting and provide a solution to problems which are related to the Frame problem. Unlike most work on the frame problem the logic described in this paper is monotonic. It (implicitly) allows for the occurrence of actions of multiple agents by introducing non-stationary notions of waiting and test. The need to state a large number of frame axioms is alleviated by introducing a concept of chronological preservation to dynamic logic. As a side effect, this concept permits the encoding of temporal properties in a natural way. We compare the relative merits of our approach and non-monotonic approaches as regards different aspects of the frame problem. Technically, we show that the resulting extended systems of propositional dynamic logic preserve (weak) completeness, finite model property and decidability.