Applications and Innovations in Intelligent Systems XIV

Restaurant table management can have significant impact on both profitability and the customer experience. The core of the issue is a complex dynamic combinatorial problem. We show how to model the problem as constraint satisfaction, with extensions which generate flexible seating plans and which maintain stability when changes occur. We describe an implemented system which provides advice to users in real time. The system is currently being evaluated in a restaurant environment.

This paper examines how data mining techniques can assist the monitoring of Crime Scene Investigator performance. The findings show that Investigators can be placed in one of four groups according to their ability to recover DNA and fingerprints from crime scenes. They also show that their ability to predict which crime scenes will yield the best opportunity of recovering forensic samples has no correlation to their actual ability to recover those samples.

Some learning theories emphasize the benefits of group work and shared knowledge acquisition in the learning processes. The Computer-Supported Collaborative Learning (CSCL) systems are used to supportcollaborative learning and knowledge building, making communication tools, shared workspaces, and automatic analysis tools available to users. In this article we describe a Bayesian network automatically built from a database of analysis indicators qualifying the individual work, the group work, and the solutions built in a CSCL environment that supports a problem solving approach. This network models the relationships between the indicators that represent both the collaborative workprocess and the problem solution.

Bayesian networks can provide a suitable framework for the integration of highly heterogeneous experimental data and domain knowledge from experts and ontologies. In addition, they can produce interpretable and understandable models for knowledge discovery within complex domains by providing knowledge of casual and other relationships in the data. We have developed a system using Bayesian Networks that enables domain experts to express their knowledge and integrate it with a variety of other sources such as protein-protein relationships and to cross-reference this against new knowledge discovered by the proteomics experiments. The underlying Bayesian mechanism enables a form of hypothesis testing and evaluation.

A collection consisting of the images of 774 live moth individuals, each moth belonging to one of 35 different UK species, was analysed to determine if data mining techniques could be used effectively for automatic species identification. Feature vectors were extracted from each of the moth images and the machine learning toolkit WEKA was used to classify the moths by species using the feature vectors. Whereas a previous analysis of this image dataset reported in the literature [1] required that each moth’s least worn wing region be highlighted manually for each image, WEKA was able to achieve a greater level of accuracy (85%) using support vector machines without manual specification of a region of interest at all. This paper describes the features that were extracted from the images, and the various experiments using different classifiers and datasets that were performed. The results show that data mining can be usefully applied to the problem of automatic species identification of live specimens in the field.

Photometry is used as a cheap and easy way to estimate redshifts of galaxies, which would otherwise require considerable amounts of expensive telescope time. However, the analysis of photometric redshift datasets is a task where it is sometimes difficultto achievea high classification accuracy. This work presents a custom Genetic Algorithm (GA) for mining the Hubble Deep Field North (HDF-N) datasets to achieve accurate IF-THEN classification rules. This kind of knowledge representation has the advantage of being intuitively comprehensible to the user, facilitating astronomers’ interpretation of discovered knowledge. The GA is tested againstthe state of the art decision tree algorithm C5.0 [Rulequest, 2005] in two datasets, achieving better classification accuracy and simplerrule sets in both datasets.

In order to optimise the energy production in a fleet of power plants, it is necessary to solve a mixed integer optimisation problem. Traditionally, the continuous parts of the problem are linearized and a Simplex scheme is applied. Alternatively, heuristic “bionic” optimisation methods can be used without having to linearize the problem. Weare going to demonstrate this approach by modelling power plant blocks with fast Neural Networks and optimising the operation of multi-block power plants over one day with Simulated Annealing.

A genetic algorithm-based procedure for solving the optimal transceiver placement problem in an optical communication broadband network is presented. It determines the minimal number and type of transceivers and their geographic distribution in an optical network in order to guarantee the broadband demanded by the end user. The proposed method has been validated for standards networks of 10, 15 and 20 nodes, and compared with classical techniques of discrete location theory. Then, the genetic algorithm has been used on a real network (70 nodes) for the West area of Malaga city (Spain).

The quality of steel strip produced in a wide strip rolling mill depends heavily on the careful selection of initial ground work roll profiles for each of the mill stands in the finishing train. In the past, these profiles were determined by human experts, based on their knowledge and experience. In previous work, the profiles were successfully optimised using a self-organising migration algorithm (SOMA). In this research, SASS, a novel heuristic optimisation algorithm that has only one control parameter, has been used to find the optimum profiles for a simulated rolling mill. The resulting strip quality produced using the profiles found by SASS is compared with results from previous work and the quality produced using the original profile specifications. The best set of profiles found by SASS clearly outperformed the original set and performed equally well as SOMA without the need of finding a suitable set of control parameters.

The ADM (Autonomous Decision Maker) concept concerns the possibility of including intelligent interfaces, agent like, for supporting the use of ubiquitous networks, such as the Internet, in safety related applications. The need for such interfaces is inevitable if remote surveillance and control shall be supported. The single most important aspect of ADM is its capability of handling limited resources when making intelligent decisions. Intelligence in ADM is manifested in reasoning and learning. This paper outlines the role of ADM and especially in relation to the standard IEC 61508 and presents the overall properties that result. These are exemplified by a presentation of ADM demonstrator.

Semantic Web technologies are emerging technologies which can considerably improve the information sharing process by overcoming the problems of current Web portals. Portals based on Semantic Web technologies represent the next generation of Web portals, however, before industry is willing to adopt Semantic Web technologies it is essential to demonstrate that Semantic Web portals are significantly better than Web portals. This paper focuses on a case study which compares the performance of a traditional Web portal using a keyword-based search engine and a Semantic Web portal using an ontology-based search engine. The empirical results of the comparison performed between these two search engines over an input data set of 100 data provides strong evidence of the tangible benefits of using Semantic Web technologies.

jCOLIBRI¹ is a Java framework that helps designing Case Based Reasoning systems. This paper presents the incorporation of Description Logics reasoning capabilities to the new release of the framework. With this extension jCOLIBRI facilitates the development of Knowledge Intensive CBR applications. Ontologies are useful regarding different aspects: as the vocabulary to describe cases and/or queries, as a knowledge structure where the cases are located, and as the knowledge source to achieve semantic reasoning methods for similarity assessment and case adaptation that are reusable across different domains.

Many combinatorial problems can be modelled as Constraint Satisfaction Problems (CSPs). Solving a general CSP is known to be NPcomplete; so that closure and heuristic search are usually used. However, many problems are inherently distributed and the problem complexity can be reduced by dividing the problem into a set of subproblems. Nevertheless, general distributed techniques are not always appropriate to distribute real life problems. In this work, we model the railway scheduling problem by means of domain dependent distributed constraint models and we show that these models maintained better behaviors than general distributed models based on graph partitioning. The evaluation is focussed on the railway scheduling problem, where domain dependent models carry out a problem distribution by means of trains and contiguous set of stations.

There is an extensive body of work on Intelligent Tutoring Systems: computer environments for education, teaching and training that adapt to the needs of the individual learner. Work on personalisation and adaptivity has included research into allowing the student user to enhance the system’s adaptivity by improving the accuracy of the underlying learner model. Open Learner Modelling, where the system’s model of the user’s knowledge is revealed to the user, has been proposed to support student reflection on their learning. Increased accuracy of the learner model can be obtained by the student and system jointly negotiating the learner model. We present the initial investigations into a system to allow people to negotiate the model of their understanding of a topic in natural language. This paper discusses the development and capabilities of both conversational agents (or chatbots) and Intelligent Tutoring Systems, in particular Open Learner Modelling. We describe a Wizard-of-Oz experiment to investigate the feasibility of using a chatbot to support negotiation, and conclude that a fusion of the two fields can lead to developing negotiation techniques for chatbots and the enhancement of the Open Learner Model. This technology, if successful, could have widespread application in schools, universities and other training scenarios.

Control over a closed domain of textual material removes many question answering issues, as does an ontology that is closely intertwined with its sources. This pragmatic, shallow approach to many challenging areas of research in adaptive hypermedia, question answering, intelligent tutoring and humancomputer interaction has been put into practice at Cambridge in the Computer Science undergraduate course to teach the hardware description language Veri/og. This language itself poses many challenges as it crosses the interdisciplinary boundary between hardware and software engineers, giving rise to severalhuman ontologies as well as theprogramming language itself We present further results from ourformal and informal surveys. We look at further work to increase the dialogue between studentand tutor and export our knowledge to the Semantic Web.

This paper introduces three grammar-segmentation methods capable of handling the large grammar issues associated with producing a real-time speech-enabled VXML bus travel application for London. Large grammars tend to produce relatively slow recognition interfaces and this work shows how this limitation can be successfully addressed. Comparative experimental results show that the novel last-word recognition based grammar segmentation method described here achieves an optimal balance between recognition rate, speed of processing and naturalness of interaction.

What is the relation between the process of adopting new technologies, and its impact on business value, in situations of high internal and external uncertainty? Whereas technology adoption is generally fairly well understood, the models do not seem to hold in situations of high uncertainty. The aim of this paper is to investigate the impact of this uncertainty, using a case study on the introduction of a new technology in a large Egyptian public bank. After exploring the most relevant uncertainty factors and their impact on the adoption process, the paper ends with a general discussion and conclusion.

RoboCup 3D Soccer Simulation is a growing domain that makes a wide variety of AI and Multi-Agent researches possible. The RoboCup 3D Soccer Simulation Server is a Multi-Agent environment that supports 22 independent agents to play a soccer match within a real-time and complex environment. Many researchers from all over the world have been using this simulator to pursue their researches in a wide variety of areas such as multiagent learning, cooperative actions and multiagent planning. This paper illustrates the current organization of RoboCup 3D Soccer Simulation Server.

Police analysts are requiredto unravel the complexities in data to assist operational personnel in arresting offenders and directing crime prevention strategies. However, the volume of crime that is being committed and the awareness of modern criminals make this a daunting task. The ability to analyse this amount of data with its inherent complexities without. using computational support puts a strain on human resources. This paper examines the current techniques that are used to predict crime and criminality. Over time, these techniques have been refined and have achieved limited success. They are concentrated into three categories: statistical methods, these mainly relate to the journey to crime, age of offending and offending behaviour; techniques using geographical information systems that identify crime hot spots, repeat victimisation, crime attractors and crime generators; a miscellaneous group which includes machine learning techniques to identify patterns in criminal behaviour and studies involving reoffending. The majority of current techniques involve the prediction of either a single offender’s criminality or a single crimetype’s next offence. These results are of only limited use in practical policing. It is our contention that Knowledge Discovery in Databases should be used on all crime types together with offender data, as a whole, to predict crime and criminality within a small geographical area of a police force.

In this paper the calibration of well known biological system namely Lotka-Volterra model is done using Genetic Algorithms. The problem of parameter estimation is formulated as an optimization problem, which is highly non linearand multimodal in nature. Binary Genetic Algorithms as well as Real Genetic Algorithms have been used to obtain the results.The comparative study showsthat the Real Genetic Algorithm is more promising.