Recursive learning method for knowledge-based planning system

In the status selection planning system, which is a kind of knowledge-based planning system, the quality of the solution depends on the status selection rules. However, it is usually difficult to acquire useful knowledge from human experts. The learning method of a status selection rule using inductive learning is proposed. The status selection rules are divided into several stages according to the planning process. Gathering a training set and learning a part of the knowledge inductively are repeated one by one from the previous stage rules. From the result of application to a job-shop problem, the effectiveness of the proposed method is shown.     

Inductive rule generation and interpretation

Abstract

The problem of knowledge acquisition has been recognized as the major bottleneck in the development of knowledge-based systems. An encouraging approach to alleviate this problem is inductive learning. Inductive learning systems accept, as input, a set of data that represent instances of the problem domain and produce, as output, the rules of the knowledge base. Each data item is described by a set of attribute values and is assigned to a unique decision class. A common characteristic of the existing inductive learning systems, is that they are empirical in nature and do not take into account the implications of the inductive rule generation process on the performance of the resulting set of rules. That performance is assessed when the rules are used to classify new unlabelled data. This paper demonstrates that the performance of a rule set is a function of the rule generation and rule interpretation processes. These two processes are interrelated and should not be considered separately. The interrelation of rule generation and rule interpretation is analysed and suggestions to improve the performance of existing inductive learning systems, are forwarded.     

Cascade ARTMAP: integrating neural computation and symbolic knowledge processing

This paper introduces a hybrid system termed cascade adaptive resonance theory mapping (ARTMAP) that incorporates symbolic knowledge into neural-network learning and recognition. Cascade ARTMAP, a generalization of fuzzy ARTMAP, represents intermediate attributes and rule cascades of rule-based knowledge explicitly and performs multistep inferencing. A rule insertion algorithm translates if-then symbolic rules into cascade ARTMAP architecture. Besides that initializing networks with prior knowledge can improve predictive accuracy and learning efficiency, the inserted symbolic knowledge can be refined and enhanced by the cascade ARTMAP learning algorithm. By preserving symbolic rule form during learning, the rules extracted from cascade ARTMAP can be compared directly with the originally inserted rules. Simulations on an animal identification problem indicate that a priori symbolic knowledge always improves system performance, especially with a small training set. Benchmark study on a DNA promoter recognition problem shows that with the added advantage of fast learning, cascade ARTMAP rule insertion and refinement algorithms produce performance superior to those of other machine learning systems and an alternative hybrid system known as knowledge-based artificial neural network (KBANN). Also, the rules extracted from cascade ARTMAP are more accurate and much cleaner than the NofM rules extracted from KBANN.     

ECA rule learning in dynamic environments

Through the development of management and intelligent control systems, we can make useful decision by using incoming data. These systems are used commonly in dynamic environments that some of which are been rule-based architectures. Event–Condition–Action (ECA) rule is one of the types that are used in dynamic environments. ECA rules have been designed for the systems that need automatic response to certain conditions or events. Changes of environmental conditions during the time are important factors impacting a reduction of the effectiveness of these rules which are implied by changing users demands of the systems that vary over time. Also, the rate of the changes in the rules are not known which means we are faced with the lack of information about rate of occurrence of new unknown conditions as a result of dynamics environments. Therefore, an intelligent rule learning is required for ECA rules to maintain the efficiency of the system. To the best knowledge of the authors, ECA rule learning has not been investigated. An intelligent rule learning for ECA rules are studied in this paper and a method is presented by using a combination of multi flexible fuzzy tree (MFlexDT) algorithm and neural network. Hence data loss could be avoided by considering the uncertainty aspect. Owing to runtime, speed, and also stream data in dynamic environments, a hierarchical learning model is proposed. We evaluate the performance of the proposed method for resource management in the Grid and e-commerce as case studies by modeling and simulating. A case study is presented to show the applicability of the proposed method.     

Concept Formation During Interactive Theory Revision

This article examines the problem of concept formation in machine learning, and focuses in particular on the problem of aggregation, i .e., the decision of which objects are to be grouped together into a new concept. While existing concept formation approaches have mainly concentrated on aggregation constraints that rely on structural or correlational properties of the concepts themselves, we argue that in an integrated learning system, other learning activities can provide an additional context that focuses concept formation before structural criteria are applied. In particular, we present the concept formation method realized by the KRT and CLT components of the integrated learning system MOBAL. In MOBAL, a concept formation attempt is triggered whenever no existing concept can adequately capture the rule instance and exception sets as they arise from the theory revision activities of the system. We describe how the so-proposed aggregate is characterized by a set of (function-free) first-order Horn clauses and how these are evaluated according to structural criteria to decide about the introduction of the concept into the representation. We show how a structural criterion can be used to ensure that any new concept improves the structure of the knowledge base, and we empirically evaluate how the introduction of new concepts according to different criteria affects the classification accuracy of learned rules.     

Study on shop floor control system in semiconductor fabrication by self-organizing map-based intelligent multi-controller

To confirm semiconductor wafer fabrication (FAB) operating characteristics, the scheduling decisions of shop floor control systems (SFCS) must develop a multiple scheduling rules (MSRs) approach in FABs. However, if a classical machine learning approach is used, an SFCS in FABs knowledge base (KB) can be developed by using the appropriate MSR strategy (this method is called an intelligent multi-controller in this study) as obtained from training examples. A classical machine learning approach main disadvantage is that the classes (scheduling decision variables) to which training examples are assigned must be pre-defined. This process becomes an intolerably time-consuming task. In addition, although the best decision rule can be determined for each scheduling decision variable, the combination of all the decision rules may not simultaneously satisfy the global objective function. To address these issues, this study proposes an intelligent multi-controller that incorporates three main mechanisms: (1) a simulation-based training example generation mechanism, (2) a data preprocessing mechanism, and (3) a self-organizing map (SOM)-based MSRs selection mechanism. These mechanisms can overcome the long training time problem of the classical machine learning approach in the training examples generation phase. Under various production performance criteria over a long period, the proposed intelligent multi-controller approach yields better system performance than fixed decision scheduling rules for each of the decision variables at the start of each production interval.     

Generating fuzzy rules for target tracking using a steady-stategenetic algorithm

Radar target tracking involves predicting the future trajectory of a target based on its past positions. This problem has been dealt with using trackers developed under various assumptions about statistical models of process and measurement noise and about target dynamics. Due to these assumptions, existing trackers are not very effective when executed in a stressful environment in which a target may maneuver, accelerate, or decelerate and its positions be inaccurately detected or missing completely from successive scans. To deal with target tracking in such an environment, recent efforts have developed fuzzy logic-based trackers. These have been shown to perform better as compared to traditional trackers. Unfortunately, however, their design may not be easier. For these trackers to perform effectively, a set of carefully chosen fuzzy rules are required. These rules are currently obtained from human experts through a time-consuming knowledge acquisition process of iterative interviewing, verifying, validating, and revalidating. To facilitate the knowledge acquisition process and ensure that the best possible set of rules be found, we propose to use an automatic rule generator that was developed based on the use of a genetic algorithm (GA). This genetic algorithm adopts a steady-state reproductive scheme and is referred to as the steady-state genetic algorithm (SSGA) in this paper. To generate fuzzy rules, we encode different rule sets in different chromosomes. Chromosome fitness is then determined according to a fitness function defined in terms of the number of track losses and the prediction accuracy when the set of rules it encodes is tested against training data. The rules encoded in the fittest chromosome at the end of the evolutionary process are taken to be the best possible set of fuzzy rules     

Comparing extended classifier system and genetic programming for financial forecasting: an empirical study

As a broad subfield of artificial intelligence, machine learning is concerned with the development of algorithms and techniques that allow computers to learn. These methods such as fuzzy logic, neural networks, support vector machines, decision trees and Bayesian learning have been applied to learn meaningful rules; however, the only drawback of these methods is that it often gets trapped into a local optimal. In contrast with machine learning methods, a genetic algorithm (GA) is guaranteeing for acquiring better results based on its natural evolution and global searching. GA has given rise to two new fields of research where global optimization is of crucial importance: genetic based machine learning (GBML) and genetic programming (GP). This article adopts the GBML technique to provide a three-phase knowledge extraction methodology, which makes continues and instant learning while integrates multiple rule sets into a centralized knowledge base. Moreover, the proposed system and GP are both applied to the theoretical and empirical experiments. Results for both approaches are presented and compared. This paper makes two important contributions: (1) it uses three criteria (accuracy, coverage, and fitness) to apply the knowledge extraction process which is very effective in selecting an optimal set of rules from a large population; (2) the experiments prove that the rule sets derived by the proposed approach are more accurate than GP.     

Scheduling dependent tasks on identical machines using a novel heuristic criterion: A robotic computation example

The problem of schedulingn dependent tasks, with arbitrary processing times, onm identical machines so as to minimize the makespan criterion is considered. Since this problem is NP-hard in the strong sense, it can be solved only suboptimally using heuristic approaches. Two new heuristic algorithms (dispatching rules), namely MVT/MISF and DMVT/MISF algorithms, for this problem are proposed. These algorithms are then used, together with the existing ones CP/MISF and DHLF/MISF, as a dispatching rule base of a new adaptively weighted combinatorial dispatching (AWCD) rule. This combinatorial dispatching rule has a superior behaviour compared to simple dispatching rules. Extended experimentation with these algorithms supports this argument. Here a representative robotic dynamics computation example is included. In addition, some empirical rules are derived and proposed for the selection of a simple dispatching rule (heuristic) if such a selection is required, for each particular input data set. These methods, as well as the existing optimal algorithms for special solvable cases of the considered problem, have been integrated in a decision support system (DSS).     

Incremental Induction of Decision Rules from Dominance-based Rough Approximations

An incremental algorithm generating satisfactory decision rules and a rule post-processing technique are presented. The rule induction algorithm is based on the Apriori algorithm. It is extended to handle preference-ordered domains of attributes (called criteria) within Variable Consistency Dominance-based Rough Set Approach. It deals, moreover, with the problem of missing values in the data set. The algorithm has been designed for medical applications which require: (i) a careful selection of the set of decision rules representing medical experience and (ii) an easy update of these decision rules because of data set evolving in time, and (iii) not only a high predictive capacity of the set of decision rules but also a thorough explanation of a proposed decision. To satisfy all these requirements, we propose an incremental algorithm for induction of a satisfactory set of decision rules and a post-processing technique on the generated set of rules. Userʼns preferences with respect to attributes are also taken into account. A measure of the quality of a decision rule is proposed. It is used to select the most interesting representatives in the final set of rules.     

Using B-spline neural network to extract fuzzy rules for a centrifugal pump monitoring

In mechanical equipment monitoring tasks, fuzzy logic theory has been applied to situations where accurate mathematical models are unavailable or too complex to be established, but there may exist some obscure, subjective and empirical knowledge about the problem under investigation. Such kind of knowledge is usually formalized as a set of fuzzy relationships (rules) on which the entire fuzzy system is based upon. Sometimes, the fuzzy rules provided by human experts are only partial and rarely complete, while a set of system input/output data are available. Under such situations, it is desirable to extract fuzzy relationships from system data and combine human knowledge and experience to form a complete and relevant set of fuzzy rules. This paper describes application of B-spline neural network to monitor centrifugal pumps. A neuro-fuzzy approach has been established for extracting a set of fuzzy relationships from observation data, where B-spline neural network is employed to learn the internal mapping relations from a set of features/conditions of the pump. A general procedure has been setup using the basic structure and learning mechanism of the network and finally, the network performance and results have been discussed.     

Application of reinforcement learning for agent-based production scheduling

Reinforcement learning (RL) has received some attention in recent years from agent-based researchers because it deals with the problem of how an autonomous agent can learn to select proper actions for achieving its goals through interacting with its environment. Although there have been several successful examples demonstrating the usefulness of RL, its application to manufacturing systems has not been fully explored yet. In this paper, Q-learning, a popular RL algorithm, is applied to a single machine dispatching rule selection problem. This paper investigates the application potential of Q-learning, a widely used RL algorithm to a dispatching rule selection problem on a single machine to determine if it can be used to enable a single machine agent to learn commonly accepted dispatching rules for three example cases in which the best dispatching rules have been previously defined. This study provided encouraging results that show the potential of RL for application to agent-based production scheduling.     

Learning fuzzy rules for similarity assessment in case-based reasoning

Fundamental to case-based reasoning is the assumption that similar problems have similar solutions. The meaning of the concept of “similarity” can vary in different situations and remains an issue. This paper proposes a novel similarity model consisting of fuzzy rules to represent the semantics and evaluation criteria for similarity. We believe that fuzzy if-then rules present a more powerful and flexible means to capture domain knowledge for utility oriented similarity modeling than traditional similarity measures based on feature weighting. Fuzzy rule-based reasoning is utilized as a case matching mechanism to determine whether and to which extent a known case in the case library is similar to a given problem in query. Further, we explain that such fuzzy rules for similarity assessment can be learned from the case library using genetic algorithms. The key to this is pair-wise comparisons of cases with known solutions in the case library such that sufficient training samples can be derived for genetic-based fuzzy rule learning. The evaluations conducted have shown the superiority of the proposed method in similarity modeling over traditional schemes as well as the feasibility of learning fuzzy similarity rules from a rather small case base while still yielding competent system performance.     

基于组合训练的规则嵌入多智能体强化学习方法

多智能体强化学习方法在仿真模拟、游戏对抗、推荐系统等许多方面取得了突出的进展。然而,现实世界的复杂问题使得强化学习方法存在无效探索多、训练速度慢、学习能力难以持续提升等问题。该研究嵌入规则的多智能体强化学习技术,提出基于组合训练的规则与学习结合的方式,分别设计融合规则的多智能体强化学习模型与规则选择模型,通过组合训练将两者有机结合,能够根据当前态势决定使用强化学习决策还是使用规则决策,有效解决在学习中使用哪些规则以及规则使用时机的问题。依托中国电子科技集团发布的多智能体对抗平台,对提出的方法进行实验分析和验证。通过与内置对手对抗,嵌入规则的方法经过约1.4万局训练就收敛到60%的胜率,而没有嵌入规则的算法需要约1.7万局的时候收敛到50%的胜率,结果表明嵌入规则的方法能够有效提升学习的收敛速度和最终效果。     

A feature selection bayesian approach for extracting classification rules with a clustering genetic algorithm

This work describes a method that combines a Bayesian feature selection approach with a clustering genetic algorithm to get classification rules in data-mining applications. A Bayesian network is generated from a data set and the Markov blanket of the class variable is applied to the feature subset selection task. The general rule extraction method is simple and consists of employing the clustering process in the examples of each class separately. In this way, clusters of similar examples are found for each class. These clusters can be viewed as subclasses and can, consequently, be modeled into logical rules. In this context, the problem of finding the optimal number of classification rules can be viewed as the problem of finding the best number of clusters. The Clustering Genetic Algorithm can find the best clustering in a data set, according to the Average Silhouette Width criterion, and it was applied to extract classification rules. The proposed methodology is illustrated by means of simulations in three data sets that are benchmarks for data-mining methods--Wisconsin Breast Cancer, Mushroom, and Congressional Voting Records. The rules extracted with all the attributes are compared to those extracted with the features belonging to the Markov blanket and the obtained results show that the proposed method is very promising.     

Reducing one class of machine learning algorithms to logical operations of plausible reasoning

The ways to transform a wide class of machine learning algorithms into processes of plausible reasoning based on known deductive and inductive rules of inference are shown. The employed approach to machine learning problems is based on the concept of a good classification (diagnostic) test for a given set of positive and negative examples. The problem of inferring all good diagnostic tests is to search for the best approximations of the given classification (partition or the partitioning) on the established set of examples. The theory of algebraic lattice is used as a mathematical language to construct algorithms of inferring good classification tests. The advantage of the algebraic lattice is that it is given both as a declarative structure, i.e., the structure for knowledge representation, and as a system of dual operations used to generate elements of this structure. In this work, algorithms of inferring good tests are decomposed into subproblems and operations that are the main rules of plausible human inductive and deductive reasoning. The process of plausible reasoning is considered as a sequence of three mental acts: implementing the rule of reasoning (inductive or deductive)with obtaining a new assertion, refining the boundaries of reasoning domain, and choosing a new rule of reasoning (deductive or inductive one).     

GA-based learning bias selection mechanism for real-time scheduling systems

The use of machine learning technologies in order to develop knowledge bases (KBs) for real-time scheduling (RTS) problems has produced encouraging results in recent researches. However, few researches focus on the manner of selecting proper learning biases in the early developing stage of the RTS system to enhance the generalization ability of the resulting KBs. The selected learning bias usually assumes a set of proper system features that are known in advance. Moreover, the machine learning algorithm for developing scheduling KBs is predetermined. The purpose of this study is to develop a genetic algorithm (GA)-based learning bias selection mechanism to determine an appropriate learning bias that includes the machine learning algorithm, feature subset, and learning parameters. Three machine learning algorithms are considered: the back propagation neural network (BPNN), C4.5 decision tree (DT) learning, and support vector machines (SVMs). The proposed GA-based learning bias selection mechanism can search the best machine learning algorithm and simultaneously determine the optimal subset of features and the learning parameters used to build the RTS system KBs. In terms of the accuracy of prediction of unseen data under various performance criteria, it also offers better generalization ability as compared to the case where the learning bias selection mechanism is not used. Furthermore, the proposed approach to build RTS system KBs can improve the system performance as compared to other classifier KBs under various performance criteria over a long period.