Applications and Innovations in Intelligent Systems XVI

In recent years IEEE 802.11 wireless local area networks (WLANs) have become increasingly popular. Consequently, there has also been a surge in the number of end-users. The IEEE 802.11 standards do not provide any mechanism for load distribution and as a result user quality of service (QoS) degrades significantly in congested networks where large numbers of users tend to congregate in the same area. The objective of this paper is to provide load balancing techniques that optimise network throughput in areas of user congestion, thereby improving user QoS. Specifically, we develop micro-genetic and standard genetic algorithm approaches for the WLAN load balancing problem, and we analyse their strengths and weaknesses. We also compare the performance of these algorithms with schemes currently in use in IEEE 802.11 WLANs. The results demonstrate that the proposed genetic algorithms give a significant improvement in performance over current techniques. We also show that this improvement is achieved without penalising any class of user.

Two neural network based vision subsystems for image recognition in micromechanics were developed. One subsystem is for shape recognition and another subsystem is for texture recognition. Information about shape and texture of the micro workpiece can be used to improve precision of both assembly and manufacturing processes. The proposed subsystems were tested off-line in two tasks. In the task of 3mm screw shape recognition the recognition rate of 92.5% was obtained for image database of screws manufactured with different positions of the cutters. In the task of texture recognition of mechanically treated metal surfaces the recognition rate of 99.8% was obtained for image database of four texture types corresponding to metal surfaces after milling, polishing with sandpaper, turning with lathe and polishing with file. We propose to combine these two subsystems to computer vision system for manufacturing of micro workpieces.

In this paper we propose a novel swallowing sound recognition technique based on the limited receptive area (LIRA) neural classifier and time-frequency decomposition. Time-frequency decomposition methods commonly used in sound recognition increase dimensionality of the signal and require steps of feature selection and extraction. Quite often feature selection is based on a set of empirically chosen statistics, making the pattern recognition dependent on the intuition and skills of the investigator. A limited set of extracted features is then presented to a classifier. The proposed method avoids the steps of feature selection and extraction by delegating them to a limited receptive area neural (LIRA) classifier. LIRA neural classifier utilizes the increase in dimensionality of the signal to create a large number of random features in the time-frequency domain that assure a good description of the signal without prior assumptions of the signal properties. Features that do not provide useful information for separation of classes do not obtain significant weights during classifier training. The proposed methodology was tested on the task of recognition of swallowing sounds with two different algorithms of time-frequency decomposition, short-time Fourier transform (STFT) and continuous wavelet transform (CWT). The experimental results suggest high efficiency and reliability of the proposed approach.

The importance of carrying out effective and sustainable agriculture is getting more and more obvious. In the past, additional fallow ground could be tilled to raise production. Nevertheless, even in industrialized countries agriculture can still improve on its overall yield. Modern technology, such as GPS-based tractors and sensor-aided fertilization, enables fanners to optimize their use of resources, economically and ecologically. However, these modern technologies create heaps of data that are not as easy to grasp and to evaluate as they have once been. Therefore, techniques or methods are required which use those data to their full capacity — clearly being a data mining task. This paper presents some experimental results on real agriculture data that aid in the first part of the data mining process: understanding and visualizing the data. We present interesting conclusions concerning fertilization strategies which result from data mining.

Image classification is usually accomplished using primitive features such as colour, shape and texture as feature vectors. Such vector model based classification has one large defect: it only deals with numerical features without considering the structural information within each image (e.g. attributes of objects, and relations between objects within one image). By including this sort of structural information, it is suggested that image classification accuracy can be improved. In this paper we introduce a framework for graph-based image classification using a weighting scheme. The schema was tested on a synthesized image dataset using different classification techniques. The experiments show that the proposed framework gives significantly better results than graph-based image classification in which no weighting is imposed.

This paper presents a software package that allows chemists to analyze spectroscopy data using innovative machine learning (ML) techniques. The package, designed for use in conjunction with lab-based spectroscopic instruments, includes features to encourage its adoption by analytical chemists, such as having an intuitive graphical user interface with a step-by-step ‘wizard’ for building new ML models, supporting standard file types and data preprocessing, and incorporating well-known standard chemometric analysis techniques as well as new ML techniques for analysis of spectra, so that users can compare their performance. The ML techniques that were developed for this application have been designed based on considerations of the defining characteristics of this problem domain, and combine high accuracy with visualization, so that users are provided with some insight into the basis for classification decisions.

Learning to rank order is a machine learning paradigm that is different to the common machine learning paradigms: learning to classify cluster or approximate. It has the potential to reveal more hidden knowledge in data than classification. Cohen, Schapire and Singer were early investigators of this problem. They took a preference-based approach where pairwise preferences were combined into a total ordering. It is however not always possible to have knowledge of pairwise preferences. In this paper we consider a distance-based approach to ordering, where the ordering of alternatives is predicted on the basis of their distances to a query. To learn such an ordering function we consider two orderings: one is the actual ordering and another one is the predicted ordering. We aim to maximise the agreement of the two orderings by varying the parameters of a distance function, resulting in a trained distance function which is taken to be the ordering function. We evaluated this work by comparing the trained distance and the untrained distance in an experiment on public data. Results show that the trained distance leads in general to a higher degree of agreement than untrained distance.

The last two years have seen the start of commercial activity within virtual worlds. Unlike computer games where Non-Player-Character avatars are common, in most virtual worlds they are the exception — and until recently in Second Life they were non-existent. However there is real commercial scope for Als in these worlds — in roles from virtual sales staff and tutors to personal assistants. Deploying an embodied AI into a virtual world offers a unique opportunity to evaluate embodied Als, and to develop them within an environment where human and computer are on almost equal terms. This paper presents an architecture being used for the deployment of chatbot driven avatars within the Second Life virtual world, looks at the challenges of deploying an AI within such a virtual world, the possible implications for the Turing Test, and identifies research directions for the future.

The paper describes several classes of tasks namely solving word problems, and classifying datasets using machine learning techniques, where the tasks may not be solvable because the information provided is incomplete. We explore the situation where one has a central concept and the missing information can either be a further descriptor / field of that concept or a (distantly) related concept. We describe how an ontology search engine has assisted in solving such problems, by summarizing the frequency of occurrence of descriptors found in a group of relevant ontologies, and by reporting which concepts are related to the central concept. The search engine used in this work has been ONTOSEARCH2. We further speculate about how such a “concept web” might be used to support the analysis and generation of natural language texts as well as spoken language.

Sharing capabilities and information between collaborating entities by using modem information- and communication-technology is a core principle in complex distributed civil or military mission scenarios. Previous work proved the suitability of Service-oriented Architectures for modelling and sharing the participating entities’ capabilities. Albeit providing a satisfactory model for capabilities sharing, pure service-orientation curtails expressiveness for information exchange as opposed to dedicated data-centric communication principles. In this paper we introduce an Information Management System which combines OWL-Ontologies and automated reasoning with Publish/Subscribe-Systems, providing for a shared but decoupled data model. While confirming existing related research results, we emphasise the novel application and lack of practical experience of using Semantic Web technologies in areas other than originally intended. That is, aiding decision support and software design in the context of a mission scenario for an unmanned system. Experiments within a complex simulation environment show the immediate benefits of a semantic information-management and -dissemination platform: Clear separation of concerns in code and data model, increased service re-usability and extensibility as well as regulation of data flow and respective system behaviour through declarative rules.

Silog is a biometrie authentication system that extends the conventional PC logon process using voice verification. Users enter their ID and password using a conventional Windows logon procedure but then the biometrie authentication stage makes a Voice over IP (VoIP) call to a VoiceXML (VXML) server. User interaction with this speech-enabled component then allows the user’s voice characteristics to be extracted as part of a simple user/system spoken dialogue. If the captured voice characteristics match those of a previously registered voice profile, then network access is granted. If no match is possible, then a potential unauthorised system access has been detected and the logon process is aborted.

The integration of Global Positioning System (GPS) technology into mobile devices provides them with an awareness of their physical location. This geospatial context can be employed in a wide range of applications including locating nearby places of interest as well as guiding emergency services to incidents. In this research, a GPS-enabled Personal Digital Assistant (PDA) is used to create a computerised tree inventory for the management of arboriculture. Using the General Packet Radio Service (GPRS), GPS information and arboreal image data are sent to a web-server. An office-based PC running customised Geographical Information Software (GIS) then automatically retrieves the GPS tagged image data for display and analysis purposes. The resulting application allows an expert user to view the condition of individual trees in greater detail than is possible using remotely sensed imagery.

This paper discusses the use of natural language or ‘conversational’ agents in e-learning environments. We describe and contrast the various applications of conversational agent technology represented in the e-learning literature, including tutors, learning companions, language practice and systems to encourage reflection. We offer two more detailed examples of conversational agents, one which provides learning support, and the other support for self-assessment. Issues and challenges for developers of conversational agent systems for e-learning are identified and discussed.

This paper presents an architecture for evolvable fuzzy rule-based classifiers, applied to the diagnosis of breast cancer, the second most frequent cause of cancer deaths in the female population. It is based on the eClass family of relative models, having the ability to evolve its fuzzy rule-base incrementally. This incremental adaptation is gradually developed by the influence that data bring, arriving from a data stream sequentially. Recent studies have shown that the eClass algorithms are very promising solution for decision making problems. Such on-line learning method has been extensively used for control applications and is also suitable for real time classification tasks, such as fault detection, diagnosis, robotic navigation etc. We propose the use of evolvable multiple-input-multipleoutput (MIMO) Takagi Sugeno Kang (TSK) rule-based classifiers of first order, to the diagnosis of breast cancer. Moreover we introduce a novel feature scoring function that identifies most valuable features of the data in real time. Our experiments show that the algorithm returns high classification rate and the results are comparable with other approaches that regard learning from numerical observations of medical nature.

Due to the complexity of nurse rostering problems (NRPs), Constraint Programming (CP) approaches on their own have shown to be ineffective in solving these highly constrained problems. We investigate a two-stage hybrid CP approach on real world benchmark NRPs. In the first stage, a constraint satisfaction model is used to generate weekly rosters consist of high quality shift sequences satisfying a subset of constraints. An iterative forward search is then adapted to extend them to build complete feasible solutions. Variable and value selection heuristics are employed to improve the efficiency. In the second stage, a simple Variable Neighborhood Search is used to quickly improve the solution obtained. The basic idea of the hybrid approach is based on the observations that high quality nurse rosters consist of high quality shift sequences. By decomposing the problems into solvable sub-problems for CP, the search space of the original problems are significantly reduced. The results on benchmark problems demonstrate the efficiency of this hybrid CP approach when compared to the state-of-the-art approaches in the literature.

In this paper we propose an evolutionary fuzzy neural networks system for extracting knowledge from a set of time series containing medical information. The series represent isokinetic curves obtained from a group of patients exercising the knee joint on an isokinetic dynamometer. The system has two parts: i) it analyses the time series input in order generate a simplified model of an isokinetic curve; ii) it applies a grammar-guided genetic program to obtain a knowledge base represented by a fuzzy neural network. Once the knowledge base has been generated, the system is able to perform knee injuries detection. The results suggest that evolved fuzzy neural networks perform better than non-evolutionary approaches and have a high accuracy rate during both the training and testing phases. Additionally, they are robust, as the system is able to self-adapt to changes in the problem without human intervention.

We present a genetic algorithm-based evolutionary computing approach to the optimisation of simulated football free kick situations. A detailed physics model is implemented in order to apply evolutionary computing techniques to the creation of strategic offensive shots and defensive player locations.

In this paper, we present an application of Artificial Intelligence (AI) to the implementation of Electronic Commerce. We provide a multi autonomous agent based framework. Our agent based architecture leads to flexible design of a spectrum of multiagent system (MAS) by distributing computation and by providing a unified interface to data and programs. Autonomous agents are intelligent enough and provide autonomy, simplicity of communication, computation, and a well developed semantics. The steps of design and implementation are discussed in depth, structure of Electronic Marketplace, an ontology, the agent model, and interaction pattern between agents is given. We have developed mechanisms for coordination between agents using a language, which is called Virtual Enterprise Modeling Language (VEML). VEML is a integration of Java and Knowledge Query and Manipulation Language (KQML). VEML provides application programmers with potential to globally develop different kinds of MAS based on their requirements and applications. We have implemented a multi autonomous agent based system called VE System. We demonstrate efficacy of our system by discussing experimental results and its salient features.

This document presents our work on a definition and experimentation of a voice interface for cultural heritage inventory. This hybrid system includes signal processing, natural language techniques and knowledge modeling for future retrieval. We discuss the first results and present some challenges for our future work.