The Handbook on Reasoning-Based Intelligent Systems

The following sections are included:

• CONTENTS

• PREFACE

This chapter presents a brief overview of the recent advances in intelligent systems. A brief summary of each chapter included in the book is presented. A list of resources is included to explore this field further.

Cognitive reasoning involves human short term memory to update the contextual information about a phenomenon. Recurrent models are often used to represent the dynamics of short term memory. These models often suffer from chaos and limit cycles. This chapter aims at designing a new approach to analyze the dynamical behavior of a cognitive reasoning system. The chapter addresses three distinct recurrent models for cognitive reasoning. The first model refers to a recurrent dynamics of a propositional temporal reasoning system. The principle of analysis to determine stable points for this model is illustrated. The second model represents a specialized cognitive map with an encoding model for weight adaptation, and a recall model for deriving new inferences. The last model considers a cognitive structure of an emotional system, the dynamics of which may suffer from limit cycles and chaos for certain parametric conditions. A method of stabilization of the system is given for the generation of stable inferences.

This chapter deals with data mining, whose aim is to extract useful information from a data set pertaining to input–output relationships of a process. A few dimensionality reduction techniques are discussed, which aim to map higher dimensional data into either 2-D or 3-D for visualization. The principles of some clustering techniques have been explained, which work based on the concept of similarity between two data points. The method of conventional regression analysis has been discussed, which can tackle the problem of forward mapping efficiently. However, it may not be always possible to carry out reverse mapping with the help of the derived regression equations. It has been felt that fuzzy logic- and neural network-based approaches can solve the problems of both forward as well as reverse mappings efficiently. Moreover, the need for design and development of an intelligent data miner has been realized recently.

Non-deterministic Information Systems (NISs) have been recognized to be the most important framework for handling the information incompleteness in tables, and some theoretical work about NISs has been investigated. We follow this robust framework, and we have been developing algorithms and utility programs for handling NISs. We are simply calling our work Rough Non-deterministic Information Analysis (RNIA). In this chapter, a framework of RNIA is briefly surveyed at first, then two significant issues on rule generation are examined. In the first issue, rule generation characterized by the consistency is picked up, and minimal certain rules and discernibility functions are introduced into NISs. In the second issue, rule generation characterized by support and accuracy criteria in NISs is investigated. An algorithm which is an extension of Apriori algorithm is also proposed. The solution of these two issues will enhance the framework of RNIA. Finally, rule generation on numerical data is considered. Here, numerical patterns with ‘@’ and ‘#’ are newly introduced into numerical values, and the equivalence classes depending upon numerical patterns are examined. The introduction of numerical patterns enhances the framework of RNIA.

We discuss the axiomatization of (theoretic) computer science within theories that include Zermelo–Fraenkel set theory. We then formulate several undecidability and incompleteness results that bear on practical situations within axiomatized versions of computer science.

This paper deals with distributed hypothesis formation in an intelligent agent system, where individual agents communicate together to overcome the incompleteness of their own observations. We present a Hypothesis Refinement framework encompassing a variety of individual hypothesis formation processes, and formalize the problem of building adequate hypotheses when data are distributed. Then, some communication protocols are proposed to let agents build proper hypotheses by confronting them to other agents in the system. At last, we give a methodology to adapt this framework to specific applications, and illustrate it with an example application in which agents attempt to flee from a building in which a fire has erupted.

We use here ‘algorithm’ in an informal way, to designate a procedure that combines mechanical steps and heuristic oracles. This algorithm is suited to deal with allocation problems. We describe it and then sketch some of its applications.

There are many systems of non-classical logics in the literature. Non-classical logics are developed to overcome several defects of classical logic. They can be regarded very important to reasoning-based intelligent systems. In this chapter, we review some important non-classical logics, in particular, modal, many-valued and paraconsistent logics. We also consider their applications to reasoning-based intelligent systems.

A paraconsistent annotated logic program called EVALPSN has been developed for dealing with defeasible deontic reasoning and applied to various kinds of intelligent control and safety verification. Moreover, in order to deal with before-after relation between processes(time intervals), bf(before-after)-EVALPSN has also been developed recently. In this chapter, bf-EVALPSN and its application to real-time process order control based on safety verification are introduced with simple examples. The bf-EVALPSN reasoning system consists of two kinds of inference rules in bf-EVALPSN called basic and transitive bf-inference rules.

This work proposes the application of preferences over abductive logic programs as an appealing declarative formalism to model choice situations. In particular, both a priori and a posteriori handling of preferences between abductive extensions of a theory are addressed as complementary and essential mechanisms in a broader framework for abductive reasoning. Furthermore, both of these choice mechanisms are combined with other formalisms for decision making, like economic decision theory, resulting in theories containing the best advantages from both qualitative and quantitative formalisms. Several examples are presented throughout to illustrate the enounced methodologies. These have been tested in our implementation, which we explain in detail.

Recursive neural networks are a powerful tool for processing structured data, thus filling the gap between connectionism, which is usually related to poorly organized data, and a great variety of real–world problems, where the information is naturally organized in entities and relationships among entities. According to the recursive paradigm, the input information consists of directed positional acyclic graphs (DPAGs), on which recursive networks are trained, following the partial order defined by the links of the graph. In this chapter, we propose a unified framework for learning in supervised neural networks, based on the BackPropagation algorithm, starting from static multi–layered architectures to recursive network. In fact, BackPropagation can be naturally specialized to train complex architectures able to process structured data, in the form of sequences and graphs. In this way, the problem of learning in connectionism, no matter what type of data is “natural” to represent a given application domain, is mantained in the elegant framework of continuous optimization. Finally, an application of the recursive model to image processing is presented, giving evidence to the fact that encoding images by trees or graphs gives rise to a more robust and informative representation, aimed at facilitating object detection and image classification.

In this expository work we present a new class of ANN — Paraconsistent Artificial Neural Networks based on Paraconsistent Annotated Evidential Logic Eτ that is capable of manipulating imprecise, contradictory and paracomplete information. Basic concepts are discussed followed with some applications in Biomedicine.

Automation and Robotics are fields that are growing very fast, both theoretically and in applications. Very imaginative theories have been proposed where classical logic is not suitable; perhaps the most famous is the so-called Fuzzy set theory, as proposed by Zadeh. This chapter presents a new non-classical logic, namely the Paraconsistent Annotated Evidential Logic Eτ, which is capable of manipulating imprecise, contradictory and paracomplete information without trivilialization. We illustrate with some applications in Automation and Robotics with some automated mobile robots showing their specific capabilities. We think that paraconsistent logical systems are very useful for applications in general opening applied techniques.

The present chapter deals with an Adaptive Intelligent Learning System (AILS) which is designed to be used with any Learning Management System (LMS). The adaptiveness provides unique method of identifying and monitoring the learner's learning processes according to their respective learning ability. AILS is a multi-agent system. This was developed in the form of JADE agents. The chapter presents the learning model, the system components, agent behavior in learner scenarios, the ontologies used in agent communications, and adaptive strategies. A sample application of AILS to a dummy LMS is also presented.

Within the design arena of modern devices based on cutting-edge processor cores, the availability of effective verification, validation and test methodologies able to automatically generate test programs represents an interesting advantage, since it can dramatically reduce the overall time for design and manufacturing, while improving yield and quality.

This chapter proposes several approaches able to automatically generate test programs for critical components of today's microprocessors. The methodologies proposed here are based on the use of an artificial tool that exploiting the simulation of biological evolution, is proven able to outperform the choices of skilled test engineers.

The experimental results presented for real complex microprocessors, show the effectiveness and the suitability of the artificial evolution to the automatic test program generation.

An agent-based model of communications traffic generated within a social network is described, which caters for various knowledge discovery techniques to be used in order to extract ‘interesting’ temporal patterns contained within anomalous data records. Attendant Java-based software — NetShow — is presented which enables analysis and display of static network configuration, links between network nodes, and correlations between the traffic passing through nominated nodes. In addition, display of network dynamics is facilitated via the incorporation of swarm techniques. Related issues of ‘similarity’, ‘familiarity’ and ‘contact lists’ are discussed within the context of Social Network Analysis and Link Mining. Finally, several suggestions are made for extending the work reported earlier in the chapter.

In this work is presented one new approach for processing the images of fingerprints, signatures, and handwritten texts, based on histogram modification and on adaptive filtration followed by image segmentation. The visual quality of the images obtained in result of the processing is significantly enhanced. The archiving is based on new method for adaptive lossless data compression, which suits the statistic of the processed images very well and ensures high compression ratio. The comparison with JPEG 2000 LS proved the method efficiency.

This chapter presents a brief introduction to logic programming and its application to knowledge representation for electronic circuits. We first show how circuits can be represented in terms of predicate logic. Basic concepts in logic programming such as facts, rules, and goals are introduced with circuits as examples. Next, we develop a grammatical method for circuit representation. Circuits are viewed as sentences, and their elements as words. Circuit structures are defined by a logic grammar called DCSG for word-order free languages. A set of grammar rules, when converted into Prolog clauses, forms a logic program which performs top-down parsing. Furthermore, we extend DCSG by introducing semantic terms into the grammar rules. These semantic terms define relationships between syntactic structures and their meanings. By assuming circuit functions to be the meanings of the circuit structures, knowledge of circuit structures and functions is coded into grammar rules. When a given circuit is parsed, not only its syntactic structure is determined, but also its functions are derived as meanings of the sentence.

The intensity of a tropical cyclone is an important factor in determining its direction of movement. This chapter presents an intelligent case-based reasoning system to help tropical cyclone forecasters predict changes in tropical cyclone intensities. Using best track data from the Joint Typhoon Warning Centre, we extract data for the coordinates, speed, direction, season and atmospheric pressure of tropical cyclones, treating each sixhour observation as a case and classifying them in three groups, “South China Sea”, “Philippines and Taiwan” and “Japan”. Similar cases are constructed by using Apriori association rule (AR) mining model, which is built-in in data mining software — SPSS Clementine 10.1, to reveal the hidden relationships between the meteorological environment and TC intensity change. Rules generated from the model are used to predict changes of intensities of TCs. Experimental results indicate that the proposed intelligent model is able to provide reasonable forecast in severe weather. In particularly, it is found that the accuracy of forecast is strongly related to the track pattern and the number of cases that are not located in the three location groups.

Standardization is one of the important preconditions to assure efficient manufacturing in industry such as in toolmaking, because the creation of standardized and hence transparent manufacturing provides the opportunity to detect and implement optimizations. In this work we describe our aim, which was to find standard processing sequences for manufacturing in toolmaking by using appropriate data mining techniques. The data which has to be analysed was collected by a special extraction, transformation and loading process from archived text files that include CAM-information.

The realized results of this analysis are that up to 75% of the procedures in manufacturing moulds can be standardized.

In this chapter, we compare the development of the artificial agents in two popular kinds of agent-based computational economic models. One is the agent-based models guided by game experiments. The other is the agent-based financial markets. While the conversation between these two classes of agent-based models is rare, the two can be connected through the idea of the generalized reinforcement learning. This is because that the artificial agents in game experiments have been well developed into a hierarchical framework such that the cognitive capacity can be incrementally added to the artificial agents from a low-level one, such as zero-intelligence agents, to a high-level one, such as belief learning agents or level-k reasoning agents. However, this hierarchy has not been found in agent-based financial markets. Therefore, bridging the two classes of agent-based models through artificial agents can help build financial agents with different level of cognitive capacity. This step is crucial for the cognitive foundation of agent-based financial markets and is related to the recent piling-up empirical studies of cognitive finance.

This chapter models the conceptual process of the classification of data for reasoning by using Tree Structure SOM and Lattice Structure SOM. We propose a new variant of structure growing SOM that puts appropriate neurons in the region on the map that has a high gradient in probability density estimation. A knowledge database is constructed automatically through data classification and rule extraction. In order to verify the validity of the proposed method, some experimental results using 2 different benchmark tests in the UCI Machine Learning Repository medical databases were described.

This paper presents an integrated environment aimed at supporting the physician in the non-invasive data analysis and medical diagnosis. Goal is to define an architectural model, in order to build an open, distributed medical diagnosis system whose quality of diagnostic feedback relies on the acquisition of medical knowledge by means of a robust theoretical approach. The model combines both fuzzy techniques for the patients' data processing and software agent-driven diagnosis, for providing a support in the interpretation of analyzed data. Moreover, the system exploits service-oriented techniques as the ground infrastructure for managing the medical knowledge and identifying clinical diagnosis and pathologies. The design of the framework constitutes the final result obtained by applying the set of guidelines which provides the worthwhile requirements in the modeling of the diagnostic process.