A parallel neurofuzzy learning and construction algorithm SPIE This paper establishes a connection between a neurofuzzy network model with the Mixture of Experts Network (MEN) modelling approach. Based on this connection, a new neurofuzzy MEN construction algorithm is proposed to overcome the curse of dimensionality that is inherent in the majority of associative memory networks and/or other rule based systems. The new construction algorithm is based on a new parallel learning method in which each model rule is trained independently, in which the parameter convergence property of the new learning method is established. By using the expert selective criterion of the MEN model output sensitivity to each expert, each rule can be selected to be trained or inhibited. The construction method is effective in overcoming the curse of dimensionality by reducing the dimensionality of the regression vector with the additional computational advantage of parallel processing. The proposed algorithm is analysed for effectiveness followed by a numerical example to illustrate the efficacy for some difficult data based modelling problem. 2001 Harris C. J. Hong X. confpaper Nonlinear model structure detection using optimum experimental design and orthogonal least squares 2001 Hong X Harris C .J confpaper A Mixture of Experts network structure construction algorithm 2001 Hong X Harris C .J confpaper Variable selection algorithm for the construction of MIMO operating point dependent neurofuzzy networks 2001 Hong X Harris C. J confpaper Neurofuzzy and SUPANOVA Modelling of Structure-Property Relationships in Al-Zn-Mg-Cu Alloys Neurofuzzy and SUPANOVA data modelling approaches have been used to determine models for yield strength and electrical conductivity from a series of experimental trials. In light of established understanding of the precipitation sequences characterising the 7xxx system, transformations of the compositional levels of important alloying elements have been derived to augment the experimental data, providing better characterisation of the main strengthening and physical characteristics of the alloys. The structure-property models identified by the neurofuzzy and SUPANOVA frameworks have been shown to lead to improvements over simple linear regression analyses, both in terms of the approximation to the experimental observations and in terms of the structure of the relationships identified. The transparency of these empirical techniques has enabled the resulting models to be validated against physical/metallurgical understanding. 2000 confpaper Polarisation-Dependent mixing in photonic crystal filled optical resonator 2000 Netti M C Harris A D Charlton M D B Parker G J Whittaker D M Baumberg J J confpaper Polarisation-dependent mixing in photonic crystal filled optical resonators 2000 Netti M C Harris A D Charlton M D B Parker G J Whittaker D M Baumberg J J confpaper Global Statistical Description of Temporal Features 2000 Shutler Jamie Nixon Mark S. Harris Chris J. confpaper The error bar estimation for the soft classification with Gaussian process models 2000 Gao J.B Gunn S.R. Harris C.J. Brown M. confpaper On a class of support vector kernels based on frames in function Hilbert spaces 2000 Gao J.B. Harris C.J. Gunn S.R. confpaper Some Remarks on Kalman Filters for the Multisensor Fusion 2000 Harris C.J. Gao J.B. confpaper Decision feedback equalizer design using support vector machines IEE 3 We consider the conventional decision feedback equalizer (DFE) that employs a linear combination of the channel observations and the past decisions. The design of this class of DFE is to construct a hyperplane that separates the different signal classes. It is well known that the popular minimum mean square error (MMSE) design is generally not the optimal minimum bit error rate (MBER) solution. We propose a strategy for designing the DFE based on support vector machines (SVM). The SVM design achieves asymptotically the MBER solution and is superior in performance to the usual MMSE solution. Unlike the exact MBER solution, this SVM solution can be computed much more efficiently. 2000 Chen S. Gunn S. Harris C. J. confpaper Statistical Gait Description via Temporal Moments 2000 Shutler Jamie D. Nixon Mark S. Harris Chris J. confpaper Statistical Gait Recognition via Velocity Moments IEE 00/018 2000 Shutler Jamie D. Nixon Mark S. Harris Chris J. confpaper A probabilistic framework for SVM regression and error bar estimation 2000 Gao J.B Gunn S.R. Harris C.J. Brown M. confpaper Adaptive Multiscale Basis Method for Modelling Discrete Nonlinear Dynamic Systems None 2000 Harris C.J. Gao J.B. confpaper Regression with input-dependent noise: a relevance vector machine treatment 2000 Gao J.B Gunn S.R. Harris C.J. Brown M. confpaper Modelling of nonlinear dynamical systems using support vector neurofuzzy networks Pergamon press-IFAC 2000 confpaper Data based constructive identification -- overcoming the curse of dimensionality Plenary paper, IFAC AIRTC symposium Budapest October 2000 The modelling of a nonlinear stochastic dynamical processes from data involves solving the problems of data gathering, preprocessing, model architecture selection, learning or adaptation, parametric evaluation and model validation. For a given model architecture such as associative memory networks, a common problem in non-linear modelling is the problem of {\it the curse of dimensionality}. A series of complementary data based constructive identification schemes, mainly based on but not limited to an operating point dependent fuzzy models, are introduced in this paper with the aim to overcome the curse of dimensionality. These include (i) a mixture of experts algorithm based on a forward constrained regression algorithm; (ii) an inherent parsimonious delaunay input space partition based piecewise local lineal modelling concept; (iii) a neurofuzzy model constructive approach based on forward orthogonal least squares and optimal experimental design and finally (iv) the neurofuzzy model construction algorithm based on basis functions that are B$\acute{e}$zier Bernstein polynomial functions and the additive decomposition. Illustrative examples demonstrate their applicability, showing that the final major hurdle in data based modelling has almost been removed. 2000 Harris C. J. Hong X. confpaper Modelling of nonlinear dynamical systems using support vector neural networks Pergamon Press 2000 confpaper Design of the optimal separating hyperplane for the decision feedback equalizer using support vector machines IEEE The conventional decision feedback equalizer (DFE) separates the different signal classes using a single hyperplane. It is well known that the popular minimum mean square error (MMSE) design is generally not the optimal minimum bit error rate (MBER) solution. We propose a method of designing the separating hyperplane for the conventional DFE based on support vector machines (SVMs). The SVM design achieves asymptotically the MBER solution and can be computed efficiently. 2000 Chen S. Harris C. J. confpaper Multisensor data fusion using Kalman filters based on neuro-fuzzy linearisation Kluwer Academic Publishers 2000 Harris C.J. Gan Q. Hyder A. K. confpaper Neurofuzzy state estimators Academic Press 2000 Harris C. J. Hong X. Gan Q. Sinha K. Gupta M. M. confpaper Neurofuzzy local linearisation modelling for state estimation of unknown nonlinear processes IOS Press 2000 Gan Q. Harris C.J. Mohammadian M. confpaper Feedback stabilization of fuzzy systems via linear matrix inequalities This paper considers the problem of designing robust feedback stabilization controller for a class of fuzzy systems. Based on piecewise Lyapunov functions, new design techniques for state feedback and output feedback controllers are proposed. The resultant controller is robust against measurement and modelling perturbations. The design conditions are non-conservative, and the design process is easy to construct by using commercially supported linear matrix inequalities software packages. 2000 Ming Feng J. Harris C. confpaper Piecewise Lyapunov Stability Conditions of Fuzzy Systems In this paper we address the stability of a class of nonlinear fuzzy systems that can be decomposed into a set of local models characterised as Takagi-Sugeno models. This new approach includes a consideration of the input membership functions, via this a reduction in the number of candidate Lyapunov functions and associated linear matrix inequalities (LMIs) is produced. This approach significantly reduces the computational load associated with determining closed loop stability as the input dimension increases. 2000 Ming Feng J. Harris C. confpaper Neurofuzzy network model construction using Bezier Bernstein polynomial functions 3 2000 Harris C. J. Hong X. confpaper A line of sight counteraction navigation algorithm for ship encounter collision avoidance 2000 Hong X Harris C. J Wilson P.A confpaper neurofuzzy mixture of experts network model construction algorithms to overcome the curse of dimensionality 2000 Harris C. J. Hong X. confpaper Neurofuzzy design and model construction of nonlinear dynamical processes from data 2000 Hong X Harris C.J confpaper Generalised neurofuzzy network modelling algorithms using Bezier Bernstein polynomial functions and additive decomposition 4 This paper introduces a new neurofuzzy model construction algorithm for nonlinear dynamic systems based upon basis functions that are Bezier Bernstein polynomial functions. This paper is generalised in that it copes with n-dimensional inputs by utilising an additive decomposition construction to overcome the curse of dimensionality associated with high n.. This new construction algorithm also introduces univariate Bezier Bernstein polynomial functions for the completeness of the generalised procedure. Like the B-spline expansion based neurofuzzy systems, Bezier Bernstein polynomial function based neurofuzzy networks hold desirable properties such as nonnegativity of the basis functions, unity of support, and interpretability of basis function as fuzzy membership functions, moreover with the additional advantages of structural parsimony and Delaunay input space partition, essentially overcoming the curse of dimensionality associated with conventional fuzzy and RBF networks. This new modelling network is based on additive decomposition approach together with two separate basis function formation approaches for both univariate and bivariate Bezier Bernstein polynomial functions used in model construction. The overall network weights are then learnt using conventional least squares methods. Numerical examples are included to demonstrate the effectiveness of this new data based modelling approach. 2000 Hong X Harris C .J confpaper A hybrid learning scheme combining EM and MASMOD algorithms for fuzzy local linearization modeling IEEE 2000 Gan Q. Harris C. J. confpaper Comparison of two measurement fusion methods for Kalman-filter-based multisensor data fusion IEEE 2000 Gan Q. Harris C. J. confpaper Adaptive Linear Finite Element Method for Modelling Nonlinear Dynamic Systems 2000 Harris C.J. Gao J.B. confpaper Linearisation and state estimation of unknown discrete-time nonlinear dynamic systems using recurrent neurofuzzy networks IEEE 6 1999 Gan Q. Harris C. J. confpaper Practical properties of a thick-film elastic wave sensor structure 1999 Harris NR White NM confpaper The application of CMAC based intelligent agents in the detection of previously unseen computer viruses This paper describes an ongoing research project at Souuthampton University into the application of the CMAC to the intelligent detection of computer viruses. 1999 Harris C J Luke J confpaper Recognising humans by gait via parametric canonical space Elsevier Science 4 Based on Principal Component Analysis (PCA), eigenspace transformation (EST) has been demonstrated to be a potent metric in automatic face recognition and gait analysis by template matching, but without using data analysis to increase classification capability. Gait is a new biometric aimed to recognise subjects by the way they walk. In this paper, we propose a new approach which combines canonical space transformation (CST) based on Canonical Analysis (CA), with eigenspace transformation for feature extraction. This method can be used to reduce data dimensionality and to optimise the class separability of different gait classes simultaneously. Each image template is projected from the high-dimensional image space to a low-dimensional canonical space. Using template matching, recognition of human gait becomes much more accurate and robust in this new space. Example results on a small database show how subjects can be recognised with 100% accuracy by their gait, using this method. 1999 Huang P.S. Harris C.J. Nixon M.S. confpaper Autonomous ship collision free trajectory navigation and control algorithms IEEE 1999 Hong X Harris C. J Wilson P.A confpaper Data Pre-Processing/Model Initialisation in Neurofuzzy Modelling of Structure-Property Relationships in Al-Zn-Mg-Cu Alloys 10 The paper deals with the application of multiple linear regression and neurofuzzy modelling approaches to 7xxx series based aluminium alloys. 36 compositional and ageing time variants and subsequent proof strength and electrical conductivity measurements have been studied. The input datasets have been transformed in two ways: to reveal more explicit microstructural information and to reflect some empirical findings in the literature. Neurofuzzy modelling exhibited improved performance in modelling proof strength and electrical conductivity c.f. the multiple linear regression approach. Electrical conductivity is best modelled using the explicit microstructural input dataset, whilst proof strength is best modelled by a further modification of this dataset, decided upon after inspection of the subnetwork structures produced by neurofuzzy modelling. Neurofuzzy modelling offers a transparent empirically based data-driven approach that can be combined with pre-processing of the data and initialising of the model structure based upon physical understanding. An iterative modelling approach is defined whereby data-driven empirical modelling approaches are first used to assess underlying data structures and are validated against physically based understanding, these then inform subsequent initialised neurofuzzy models and input data transformations to provide both optimal subset and feature representation. 1999 confpaper Recursive Bayesian Modelling of Time Series by Neural Networks 1999 Dodd T.J. Harris C.J. confpaper Neurofuzzy adaptive modelling of nonlinear dynamical systems based on multiscale function space decomposition University of Southampton 1 1999 Harris C.J. Gao J.B. confpaper Committees of Gaussian kernel based models 1999 Dodd T.J. Harris C.J. confpaper A New Multiple Model Framework for Recursive Bayesian Modelling of Time Series by Neural Networks 1999 Dodd T.J. Harris C.J. confpaper Using Hierarchical Classification to Exploit Context in Pattern Classification for Information Fusion In data fusion applications it is important that only the minimum set of relevant features are combined at any one stage in the fusion process. A hierarchical classification methodology is described which handles features at different levels of abstraction to produce a more robust and interpretable classifier. This is achieved by dividing the classes into contextual subgroups, which are further divided to produce a tree structure defining relationships between classes. A novel approach is proposed for the class structure design which is formulated as a constrained search in the structure space. This can be performed via a forward search algorithm driven by a cost function dependent on the performance of the class structure and constraints on the required solution. 1999 Bailey Alex Harris Chris confpaper Adaptive Multiscale Basis Method for Nonlinear Modelling: Function Space Decomposition Approach 1999 Harris C.J. Gao J.B. confpaper High temperature superconducting power transformers: conclusions from a design study 1 1999 Sykulski J.K. Beduz C. Stoll R.L. Harris M.R. Goddard K. Yang Y. confpaper An intelligent guidance and control system for ship obstacle avoidance ICSC An intelligent guidance and control system using a neurofuzzy network multistep ahead predictor model is introduced and applied to ship obstacle avoidance, which uses only observed input/output data generated by on board and external sensors, and a data fusion algorithm to generate the desired waypoints. A simple and effective waypoint guidance scheme based on line-of-sight is derived for a data based ship model. A neurofuzzy network predictor, based on using rudder deflection angle for the control of ship heading angle, is utilised on the simulation of ESSO 190000dwt tanker model to demonstrate the effectiveness of the system. The approach is generic and extendable to aircraft and missile control and guidance problems where the vehicle dynamics change significantly during flight in a manner dependent upon operational use, the only requirement for implementation being observed data to construct sensor and vehicle models. 1999 Harris C. J. Hong X. Wilson P.A confpaper Optimal Piecewise locally linear modelling SPIE 1999 Harris C. J. Hong X. Feng M confpaper Automatic Gait Recognition via Statistical Approaches 1999 Huang Ping Sheng Nixon Mark S. Harris Chris J. confpaper Hybrid Computed Order Tracking 4 Virbration analysis is an integral part of modern condition monitoring and fault diagnosis for rotating machinery.In this paper three different synchronous sampling schemes are developed for the run down of gas turbine shaft with faults.the results are shown to be superior to conventional methods. 1999 confpaper A new partitioning approach for constructing fuzzy models 3rd World Multiconference on Systemics, Cybernetic and Informatics (SCI'99) and 5th International Conference on Information Systems Analysis and Synthesis (ISAS'99) This paper presents a technique for constructing data based fuzzy model of a dynamical system by a new method of partitioning the data input space. The method is able to derive a fuzzy model from data automatically and completely avoids the curse of dimension problem usually associated with fuzzy system. 1999 Ming Feng J. Harris C. confpaper Neurofuzzy state estimators using a modified ASMOD and Kalman filter algorithm 1999 Gan Q. Harris C. J. confpaper Multisensor data fusion using Kalman filters based on linearised process models 1999 Gan Q. Harris C.J. confpaper Local linearisation using fuzzy (soft) input space partition for nonlinear system modelling 1999 Harris C. J. Gan Q. confpaper A multi-spectral data-fusion approach to speaker recognition 1999 Higgins J.E. Damper R.I. Harris C.J. confpaper Human Gait Recognition in Canonical Space Using Temporal Templates IEE 2 This paper describes a system for automatic gait recognition without segmentation of particular body parts. Eigenspace transformation (EST) has already proved useful for several tasks including face recognition, gait analysis, etc. However, EST is optimal in dimensionality reduction by maximising the total scatter of all classes but is not optimal for class separability. In this paper, a statistical approach which combines EST with canonical space transformation (CST) is proposed for gait recognition using temporal templates from a gait sequence as features. This method can be used to reduce data dimensionality and to optimise the class separability of different gait sequences simultaneously. Incorporating temporal information from optical-flow changes between two consecutive spatial templates, each temporal template extracted from computation of optical flow is projected from a high-dimensional image space to a single point in a low-dimensional canonical space. Using template matching, recognition of human gait becomes much faster and simpler in this new space. As such, the combination of EST and CST is shown to be of considerable potential in an emerging new biometric. 1999 Huang P.S. Harris C.J. Nixon Mark S. confpaper Neurofuzzy state identification using prefiltering 2 1999 Hong X Harris C. J Wilson P.A confpaper Neurofuzzy State Estimators and their Applications 1 1999 Harris C. J. Wu Z. Q. Gan Q. confpaper Neurofuzzy identification of an autonomous underwater vehicle 9 1999 Bossley K M Brown M Harris C J confpaper An intelligent guidance and control system for ship obstacle avoidance 14 An intelligent guidance and control system using a neurofuzzy network multistep ahead predictor model is introduced and applied to ship obstacle avoidance, which uses only observed input/output data generated by on board and external sensors, and a data fusion algorithm to generate the desired waypoints. A simple and effective waypoint guidance scheme based on line-of-sight is derived for a data based ship model. A neurofuzzy network predictor, based on using rudder deflection angle for the control of ship heading angle, is utilised on the simulation of ESSO 190000dwt tanker model to demonstrate the effectiveness of the system. The approach is generic and extendable to aircraft and missile control and guidance problems where the vehicle dynamics change significantly during flight in a manner dependent upon operational use, the only requirement for implementation being observed data to construct sensor and vehicle models. 1999 Harris C. J. Hong X. Wilson P.A confpaper Neurofuzzy local linearisation for model identification and state estimation of unknown nonlinear processes 4 1999 Gan Q. Harris C. J. confpaper Fuzzy local linearisation and local basis function expansion in nonlinear system modelling 4 1999 Gan Q. Harris C. J. confpaper VisualSurveillance and Tracking of Humans by Face and Gait Recognition Increased emphasis on automated real time intelligent surveillance system has led to the need to identify and track people in complex environments. Independent features such as face, gait provide valuable clues as to identity, which coupled with data fusion and tracking algorithms offer a potential solution to this problem. In this paper we address the first problem of recognizing humans in real time, data fusion and tracking will be performed by neurofuzzy state estimators. A new approach which combines eigenspace transformation with canonical space transformation is proposed here. This method can be used to reduce data dimensionality and to optimize the class separability of different classes simultaneously. 1998 Huang P.S. Harris C.J. Nixon M.S. confpaper Recognizing Humans by Gait using a Statistical Approach forTemporal Templates In this paper, we propose a new approach which combines canonical space transformation (CST) with the eigenspace transformation (EST) for feature extraction of temporal templates in a gait sequence. Eigenspace transformation has been demonstrated to be a potent metric in automatic face recognition and gait analysis, but without using data analysis to increase classification capability. Our method can be used to reduce data dimensionality and to optimize the class separability of different gait sequences simultaneously. Each temporal template is projected from high-dimensional image space to a single point in low-dimensional canonical space. In this new space the recognition of human gait by template matching becomes much faster and simpler. Experimental results for human gait analysis show this method is superior to eigenspace representation. As such, the combination of EST and CST is shown to be of considerable advantage in an emerging new biometric. 1998 Huang P.S. Harris C.J. Nixon M.S. confpaper A Statistical Approach for Recognizing Humans by Gait using Spatial-Temporal Templates In order to tackle the problem of recognizing humans by gait, we use an approach which combines eigenspace transformation (EST) with canonical space transformation (CST) for feature extraction of spatial templates from a gait sequence. Our proposed method can be used to reduce data dimensionality and to optimize the class separability of different gait sequences simultaneously. In this paper, we propose a new feature - temporal templates , and an extended feature which combines spatial and temporal templates for recognition. By incorporating spatial and temporal information into an extended feature vector in the canonical space, gait recognition becomes more robust and accurate than using any single feature alone. 1998 Huang P.S. Harris C.J. Nixon M.S. confpaper A Data Driven Approach to Sensor Modelling, Estimation, Tracking and Data Fusion 1998 Dodd T.J. Bailey A. Harris C.J. Bedworth M. O'Brien J. confpaper Piezoelectric thick-film polymer pastes Institute of Physics Publishing 1998 Papakostas T Harris NR Beeby SP White NM White NM confpaper Practical properties of a thick-film elastic wave sensor structure Institute of Physics Publishing 1998 Harris NR White NM White NM confpaper Comparing Different Template Features for Recognizing People by Their Gait BMVA To recognize people by their gait from a sequence of images, we have proposed a statistical approach which combined eigenspace transformation (EST) with canonical space transformation (CST) for feature transformation of spatial templates. This approach is used to reduce data dimensionality and to optimize the class separability of different gait sequences simultaneously. Good recognition rates have been achieved. Here, we incorporate temporal information from optical flows into three kinds of temporal templates and use them as features for gait recognition in addition to the spatial templates. The recognition performance for four kinds of template features has been evaluated in this paper. Experimental results show that spatial templates, horizontal-flow templates and the combined horizontal-flow and vertical-flow templates are better than vertical-flow templates for gait recognition. 1998 Huang P.S. Harris C.J. Nixon M.S. confpaper Thick-film printing of PZT onto silicon for micromechanical applications 1998 Harris N R Beeby S P White N M Koch M Evans A G R confpaper Anhur: Neuro-fuzzy Systems for Command and Control DoD C4ISR, CCRP The subject of this paper is the investigation of neuro-fuzzy modeling and control techniques as an aid to decision making in Command and Control (C2). Researchers in the military Distributed Interactive Simulation (DIS) community are exploring the possibility of using neuro-fuzzy systems to govern the decision-making behaviour of Computer Generated Forces (CGFs), autonomous agents which act as realistic participants in virtual reality training exercises. In order to provide an operational and practical means of studying neuro-fuzzy systems for C2, two elements of software are combined. The first is a simulation package, a game called Xtank, which allows teams of autonomous intelligent agents and human users to participate in simulated military operations. The second element is the Neuro-fuzzy Toolbox, which allows neuro-fuzzy network models to be constructed from system observations and to be used as rule based decision makers. Following a methodology called SOCIAL, a user of the system can configure and run wargame scenarios, systematically record data, construct neuro-fuzzy models of C2 decision-making, and then incorporate these as behavioural controllers expressed as either decision surfaces or rulebases. This paper describes some illustrative experiments and results involving Tactical Situation Assessment and discusses preliminary results in the area of Mission Evaluation. 1998 Hall Michael Harris Chris confpaper Canonical Space Representation for Recognizing Humans by Gait and Face Eigenspace transformation (EST) based on Principal Component Analysis (PCA) has been demonstrated to be a potent metric in face recognition and gait analysis, but without using data analysis to increase classification capability. In this paper, we propose a new approach which combines canonical space transformation (CST) based on Canonical Analysis (CA) with the eigenspace transformation for feature extraction. This method can be used to reduce data dimensionality and to optimise the class separability of different gait sequences and face classes simultaneously. Each image template is projected from high-dimensional image space to a single point in low-dimensional canonical space. In this new space the recognition of human gait and faces becomes much simpler. Experimental results for human gait analysis and face recognition show this method is superior to use EST or CST alone. As such, the combination of PCA and CA is shown to be of considerable advantage in an emerging new biometric. 1998 Huang P.S. Harris C.J. Nixon M.S. confpaper Webs of Research: Putting the User in Control For over 40 years the citation databases from the Institute for Scientific Information (ISI) have provided a unique and incisive tool not just for researching the academic literature but for measuring and evaluating it as well. With the emergence of the World Wide Web this tool has been developed to place cited reference searching directly within the control of the user. Improved access to information is one feature of the Web, in particular through the mechanism of the Web link. The convergence of citation searching and Web linking towards citation linking will be examined in the paper with reference to two examples: Web of Science, a new service for scientists and social scientists from ISI, and an Open Journal of Cognitive Science developed by the Open Journal project. Web of Science provides links to the literature, links to holdings, to document delivery, and integration with bibliographic tools. Using citations -- the links made by authors themselves -- users can navigate between their current work and a priori work in the archives of the research literature or take a recent paper and move forward, tracking the citations dynamically. Literature searching is incomplete without links to the primary journal literature. The Open Journal of Cognitive Science, a collaborative work between the project and ISI, illustrates how the Web model of linking has been extended, linking from the abstracted literature to the full-text journal. Both implementations are illustrated in the paper. 1998 Hitchcock Steve M. Harris Stephen W. Carr Les A. Hall Wendy confpaper Recognising Humans by Gait via Parametric Canonical Space Eigenspace transformation (EST) based on Principal Component Analysis (PCA) has been demonstrated to be a potent metric in gait analysis, but without using data analysis to increase classification capability. In this paper, we propose a new approach which combines canonical space transformation (CST) based on Canonical Analysis (CA), with the eigenspace transformation. This method can be used to reduce data dimensionality and to optimise the class separability of different gait sequences and face classes simultaneously. Each image template is projected from high-dimensional image space to a single point in low-dimensional canonical space. In this new space the recognition of human gait becomes much simpler. Experimental results for human gait analysis show this method is superior to the eigenspace representation. The comparison of EST, CST and our approach is also shown in the results. As such, the combination of EST and CST is shown to be of considerable advantage in an emerging new biometric. 1998 Huang P.S. Harris C.J. Nixon M.S. confpaper A novel micromachined pump based on thick-film piezoelectric actuation 1998 Koch M Harris N Evans A G R White N M Brunnschweiler A confpaper Wavelet Packet Analysis in the Condition Monitoring of Rotating Machinery 1998 Bossley K M McKendrick R J Harris C J Mercer C confpaper Nonlinear neurofuzzy state estimators for feature tracking and data fusion 1998 Harris C.J. Wu Z.-Q. Gan Q. confpaper Linking electronic journals: Lessons from the Open Journal project December The Open Journal project has completed its three year period of funding by the UK Electronic Libraries(eLib) programme. During that time, the number of journals that are available electronically leapt from a few tens to a few thousand. Some of these journals are now developing the sort of features the project has been advocating, in particular the use of links within journals, between different primary journals, with secondary journals data, and to non-journal sources. Assessing the achievements of the project and considering some of the difficulties it faced, we report on the different approaches to linking that the project developed, and summarise the important user responses that indicate what works and what does not. Looking ahead, there are signs of change, not just to simple linking within journals but to schemes in which links are the basis of "distributed" journals, where information may be shared and documents built from different sources. The significance has yet to be appreciated, but this would be a major change from printed journals. If projects such as this and others have provided the initial impetus, the motivation for distributed journals comes, perhaps surprisingly, from within certain parts of the industry, as the paper shows. 1998 Hitchcock Steve M. Carr Les A. Hall Wendy Harris Stephen W. confpaper Multi-Sensor Data Fusion in Defence and Aerospace 1015 1998 Harris C.J. Bailey A. Dodd T.J. confpaper Adaptive Neurofuzzy Control for A Class of State Dependent Nonlinear Processes 7 This paper presents a neurofuzzy based scheme for modeling and control of a class of nonlinear systems with an ARMA like model (a generalised Takagi-Sugeno fuzzy model), whose parameters are unknown nonlinear functions of the input and output variables or states of the plant. An associative memory network is used to identify each nonlinear function. The controller is a feedback linearising control law which can decouple the nonlinearity of the system. For the cases of adaptive and the fixed model parameters, detailed closed-loop stability analysis is carried out. It is shown that the consequent closed-loop system is globally stable. The main assumptions placed on the system and model for stability are minimum phase and a limit on the modeling mismatch error or uncertainty. Simulation examples are given to illustrate the efficacy of the proposed approach. 1998 Feng M. Harris C.J. confpaper Formulation of a screen printable piezoelectric thick-film Institute of Physics Publishing 1997 Dargie PG Harris NR Atkinson JK White NM Augousti AT White NM confpaper Hybrid delay line sensors Institute of Physics Publishing 1997 Harris NR White NM Augousti AT White NM confpaper A probabilistic framework for understanding nonlinear dynamical systems in the Hilbert space This paper deals with the probabilistic description of a general {\em a priori} unknown linear/nonlinear dynamical physical system in an observer related coordinate system. The example of a point mass interacting with its environment and the corresponding relationship to the description of an observer are used as motivation. A probabilistic formulation follows, leading to a continuity equation for probability distributions over time. The last section then demonstrates how probability densities of the spatial coordinates of the point mass may be decomposed into functions which lead to probability distributions of the velocity of the point mass. 1997 Schilhabel T.E. Harris C.J. confpaper Adaptive Identification and State Estimation of unknown nonlinear Stochastic Systems by Recurrent Neuralfuzzy Networks In this paper, the authors utilise the neural network technique and the Kalman filter algorithm to achieve adaptive and optimal modelling and filtering for unknown observable nonlinear stochastic processes. A special class of state space model is imposed on the input-output observable nonlinear stochastic system, which can be identified by a recurrent neurofuzzy network. This model enables the conventional Kalman filter to estimate the system state, avoiding the convergent problem associated with the extended Kalman filter. The training algorithm and the Kalman filter are executed interactively, thus adaptive modelling and filtering is achieved. A simulated example is given to illustrate the approach. 1997 Wu Z.Q. Harris C.J. confpaper Citation Linking: Improving Access to Online Journals The most innovative online journals are maturing rapidly and distinctive new features are emerging. Foremost among these features is the hypertext link, popularised by the World Wide Web and which will form the basis of a new, highly integrated scholarly literature. Journal integration in this instance seeks to recognise, extend and exploit relationships at the level of journal content-the papers-while maintaining some of the familiar contexts, in some cases journal identities, that define the content hierarchy and inform decision-making by readers. Links are a powerful tool for journal integration, most immediately in the form of citation linking. The paper reviews examples of citation linking in practice, and describes a new system, a link service, which is being developed to support novel and flexible linking mechanisms on the Web. One application of this link service is the Open Journal project, which is working with journal publishers to investigate the most effective ways of applying these powerful link types to enhance online journals. 1997 Hitchcock Steve M. Carr Les A. Harris Stephen W. Hey Jessie M. N. Hall Wendy confpaper A novel micromachined pump based on thick-film piezoelectric actuation 1997 Koch M Harris NR Evans AGR White NM Brunnschweiler A confpaper Overcoming the curse of dimensionality of fuzzy logic 1997 Harris C.J. confpaper Multi-sensor data fusion for situational assessment: a critical element of systems integration, some theory and application to collision avoidance 1997 Harris C.J. Doyle R.S. confpaper A 10 Year Perspective on the Theory and Application of Intelligent Modelling, Control and Estimation 1997 Harris C.J. confpaper Neurofuzzy Modelling Approaches in System Identification System identification is the task of constructing representative models of processes and has become an invaluable tool in many different areas of science and engineering. Due to the inherent complexity of many real world systems the application of traditional techniques is limited. In such instances more sophisticated (so called intelligent) modelling approaches are required. Neurofuzzy modelling is one such technique, which by integrating the attributes of fuzzy systems and neural networks is ideally suited to system identification. This attractive paradigm combines the well established learning techniques of a particular form of neural network i.e. generalised linear models with the transparent knowledge representation of fuzzy systems, thus producing models which possess the ability to learn from real world observations and whose behaviour can be described naturally as a series of linguistic humanly understandable rules. Unfortunately, the application of these systems is limited to low dimensional problems for which good quality expert knowledge and data are available. The work described in this thesis addresses this fundamental problem with neurofuzzy modelling, as a result algorithms which are less sensitive to the quality of the a priori knowledge and empirical data are developed. The true modelling capabilities of any strategy is heavily reliant on the model's structure, and hence an important (arguably the most important) task is structure identification. Also, due to the curse of dimensionality, in high dimensional problems the size of conventional neurofuzzy models gets prohibitively large. These issues are tackled by the development of automatic neurofuzzy model identification algorithms, which exploit the available expert knowledge and empirical data. To alleviate problems associated with the curse of dimensionality, aid model generalisation and enhance model transparency, parsimonious models are identified. This is achieved by the application of additive and multiplicative neurofuzzy models which exploit structural redundancies found in conventional systems. The developed construction algorithms successfully identify parsimonious models, but as a result of noisy and poorly distributed empirical data, these models can still generalise inadequately. This problem is addressed by the application of Bayesian inferencing techniques; a form of regularisation. Smooth model outputs are assumed and superfluous model parameters are controlled, sufficiently aiding model generalisation and transparency, and data interpolation and extrapolation. By exploiting the structural decomposition of the identified neurofuzzy models, an efficient local method of regularisation is developed. All the methods introduced in this thesis are illustrated on many different examples, including simulated time series, complex functional equations, and multi-dimensional dynamical systems. For many of these problems conventional neurofuzzy modelling is unsuitable, and the developed techniques have extended the range of problems to which neurofuzzy modelling can successfully be applied. 1997 Bossley K.M. Harris C.J. confpaper A Helicopter Obstacle Avoidance System Incorporating Non-linear Neurofuzzy Multi-Sensor Data Fusion Hazardous weather conditions significantly limit the operational capability of civil helicopters. This limitation arises from the crew's inability to determine the location of obstacles in the environment by sight. In order to assist the crew in these circumstances a range of equipment and sensors may be installed in the helicopter. However, with multiple sensors on board, the problem of efficiently assimilating the large amount of imagery and data available generates a significant workload. A reduction of the workload may be achieved by the automation of this assimilation (sensor fusion) and the design of a system to guide the pilot along obstacle free paths. In order to provide the guidance to avoid obstacles a system must have knowledge about the obstacles' possible positions and likely future positions relative the system's own aircraft. Since the information being provided by the sensors will not be perfect, (i.e. it will have some uncertainty associated with it), and since the process model, which must be used to predict any future positions, will also be uncertain, the required positions must be estimated. As the dynamics of moving obstacles will be a priori unknown, it will be necessary to learn process models for them. The dynamics of the obstacles cannot be guaranteed to be linear, therefore these process models must be capable of reflecting this non-linear behaviour. The uncertain information produced by the various sensors will be related to required knowledge about the obstacles by a sensor model, however this relationship need not be linear, and may even have to be learned. Currently used estimation techniques (e.g. the ordinary extended Kalman filter) are inadequate for estimating the uncertainty involved in the obstacles' positions for the highly non-linear processes under consideration. Neural network approaches to non-linear estimation have recently allowed process and sensor models to be learned (sometimes implicitly), however these approaches have been quite ad hoc in their implementation and have been even more negligent in the estimation of uncertainty. The main contributions of this research are the design of non-linear estimators which may use process and sensor models that result from learning processes, and the use of the output of these estimators to determine guidance for obstacle free paths through the environment in 3 dimensions. 1997 Doyle R.S. Harris C.J. confpaper Screen printing of thick piezoelectric PZT layers onto silicon IEE Colliquium on Recent Advances in micromachining techniques 1997 Koch M Evans A G R Brunnschweiler A Harris N White N M confpaper 1997 Koch M Harris N Maas R Evans A G R White N M Brunnschweiler A confpaper A novel micromachined pump based on thick-film piezolelectric actuation IEE Transducers 1997 Koch M Harris N Evans A G R White N M Brunnschweiler A confpaper Thick -fil printing of PZT onto silicon 1997 Maas R Koch M Harris N White N M Evans A G R confpaper Neurofuzzy Systems Modelling: A Transparent Approach Springer Verlag This chapter advocates a cyclic construction approach to data modelling based on a design-train-validate-interpret cycle. Traditional approaches to data modelling with neural networks typically produce opaque systems which are difficult to interpret and hence validate. Neurofuzzy systems equip neural networks with a linguistic interpretation which provides the designer with enhanced transparency enabling the loop to be closed in the modelling cycle. Three neurofuzzy construction algorithms are discussed, and their performances are evaluated on two real data sets. 1997 Brown M. Gunn S.R. Ng C.Y. Harris C.J. Warwick K. confpaper Intelligent Neurofuzzy Estimators and Multisensor Data Fusion Kluwer Academic Multi-Sensor Data Fusion (MSDF), or the process of fusing data from a variety of disparate data sources about a single entity, feature or system state, is of prime importance in the monitoring and control of complex systems. This paper addresses the basic subproblems of MSDF within a unified informational framework derived via Neurofuzzy modelling and estimation algorithms. This environment provides a common framework for integrating information which is database, sensor based, experiemental based and mechanistic. The paper introduces parsimonious neurofuzzy modelling algorithms and utilises them in generating local state estimators which are optimal in information processing. A MSDF system utilises these algorithms in a distributed decentralised architecture. 1997 Harris C.J. Wu Z.Q. Bossley K.M. Brown M. Tzafestas S.G. confpaper Neurofuzzy State Estimators and Their Applictions Neurofuzzy algorithms have been extensively developed in recent years for the real time/online identification of nonlinear a priori unknown dynamical processes. As with all rule base paradigms they suffer from the curse of dimensionality, restricting their practical use to low dimensional control problems. This paper shows how adaptive construction algorithms based on additive and extended additive decomposition techniques can overcome this problem, to produce parsimonious neurofuzzy models which retain their transparency or interpretability. Not only does this approach extend the applicability of neurofuzzy algorithms, it also enables low complexity controllers, estimators to be derived. In this context neurofuzzy state estimators are derived which automatically parameterise a Kalman filter for a process state estimate reconstruction from any input/output data source. This approach avoids pitfalls of the extended Kalman filter, and is optimal for local models. The paper discusses real world applications of this new theory of modelling and estimation to helicopter guidance, intelligent driver warning system, communication antennas, autonomous underwater vehicles and an IFAC benchmark problem. 1997 Harris C.J. Wu Z.Q confpaper Aspects of the Theory and Application of Intelligent Modelling, Control and Estimation Neurofuzzy algorithms have been extensively developed in recent years for the real time/online identification of nonlinear a priori unknown dynamical processes. As with all rule base paradigms they suffer from the curse of dimensionality, restricting their practical use to low dimensional control type problems. This paper shows how adaptive construction algorithms based on additive and extended additive decomposition techniques can overcome this problem, to produce parsimonious neurofuzzy models which retain their transparency or interpretability. Not only does this approach extend the applicability of neurofuzzy algorithms, it also enables low complexity controllers, or estimators to be derived. In this context neurofuzzy state estimators are derived which automatically parameterise a Kalman filter for a process state estimate reconstruction from any input/output data source. This approach avoids the usual pitfalls of the extended Kalman filter, and is optimal for local models. The local modelling approach is shown to be directly applicable to adaptive control of a priori unknown nonlinear systems. 1997 Harris C.J. Wu Z.Q. Feng M. confpaper A Real Time Neurofuzzy Modelling and State Estimation Scheme ISO Press The authors of this paper analyse the input-output relation of the fuzzy system with a functional rule base and B-spline basis functions as membership functions, constructing a kind of neurofuzzy network for system modelling with a simple but effective training algorithm. This model is applied to state estimation, a simulated example is given. 1997 Wu Z.Q. Harris C.J. Morabito F.C. confpaper A novel micropump design with thick-film piezoelectric actuation 1 1997 Koch M Harris NR Maas R Evans AGR White NM Brunnschweiler A confpaper A Neurofuzzy Network Structure for Modelling and State Estimation of Unknown Nonlinear Systems 4 A Fuzzy logic system has been shown to be able to arbitrarily approximate any nonlinear function and has been successfully applied to system modelling. The functional rule fuzzy system enables the input-output relation of the fuzzy logic system to be analysed. B-spline basis functions have many desirable numerical properties and as such can be used as membership functions of fuzzy system. This paper analyses the input-output relation of a fuzzy system with a functional rule base and B-spline basis functions as membership functions; constructing a neurofuzzy network for systems representation in which the training algorithm for this network structure is very simple since the network is linear in the weights. It is also desired to merge the neural network identification technique and the Kalman filter to achieve optimal adaptive filtering and prediction for unknown but observable nonlinear processes. In this paper, the derived neurofuzzy network is applied to state estimation in which the system model identified is converted to its equivalent state-space representation with which a Kalman filter is applied to perform state estimation. Two approaches that combine the neurofuzzy modelling and the Kalman filter algorithm, the indirect method and direct method, are presented. A simulated example is also given to illustrate the approaches based on real data. 1997 Wu Z.Q. Harris C.J. confpaper The Paper House of Cards (And Why It Is Taking So Long To Collapse) The remaining cost of serial publication, once expenses are scaled down to the electronic-only level, is low enough to render the interests of everyone -- the author, the reader, the funder of the author's research, the university supporting the author, and, yes, the electronic learned serial publishers -- better served by recovering those costs and a fair profit at the author's end, in the form of page charges (paid for by the funders of the author's research and/or the university employing him to do the research, both co-beneficiaries, with the author, of the widest possible unimpeded distribution of the research reported), rather than by any version of reader-end payment, the latter depending as it does, on restricting access to what the author and his supporters would all prefer to see as free for all. 1997 Harnad Stevan confpaper Review of Roy Harris, Signs of Writing (Routledge) 2 1997 Hughes Rebecca O'Hara Kieron confpaper Development and Application of Adaptive Neurofuzzy Modelling and Estimation in Drug Assays 1996 Harris C.J. confpaper Multi Sensor Data Fusion for Real Time Aircraft Collision Avoidance IEE 1996 Harris C.J. Doyle R.S. confpaper Intelligent Estimatiors with Applications in Multi-Data Fusion 1996 Harris C.J. confpaper Collision Avoidance in Helicopters via Neurofuzzy Multisensor Data Fusion 1996 Harris C.J. confpaper Adaptive Neurofuzzy Kalman Filter It is of great practical significance to merge the neural network identification technique and the Kalman filter to achieve adaptive and optimal filtering and prediction for unknown observable nonlinear processes. In this paper, an operating point dependent ARMA model is used to represent the nonlinear system, and a neurofuzzy network is used to approximate each AR parameter of such a model which can then be converted to its equivalent state-space representation. Using this state-space form, a Kalman filter can be applied to estimate the system state. The system modelling algorithm and the Kalman filter are combined in a bootstrap scheme, in which the error between the measured output and the filtered output is used to train the neural network, thus adaptive filtering for noisy nonlinear system is achieved. A simulated example is also given. 1996 Wu Z.Q. Harris C.J. confpaper Neurofuzzy Modelling and State Estimation It is of great practical significance to merge the neural network identification technique and the Kalman filter to achieve optimal adaptive filtering and prediction for unknown observable nonlinear processes. In this paper, an operating point dependent ARMA model is used to represent the nonlinear system, and a neurofuzzy network is used to identify this model. It is then converted to its equivalent state-space representation with which a Kalman filter is applied to perform state estimation. Two approaches to combine the neurofuzzy modelling and the Kalman filter algorithm, indirect method and direct method, are presented. A simulated example is also given. 1996 Wu Z.Q. Harris C.J. confpaper Fuzzy Logic Based Estimators and Predictors for Agile Target Tracking Applications 1996 Moore C.G. Rogers E. Harris C.J. confpaper Multi-Sensor Data Fusion for Helicopter Guidance using Neuro-Fuzzy Estimation Algorithms The purpose of this paper is to describe an approach that performs data fusion on the output of multiple spatially separate sensors engaged in the real time tracking of obstacles in a helicopter's environment. The generated information can be used either as a flight director aid or as feedback required by an automatic collision avoidance system. Obstacle track estimation has been commonly carried out using the Kalman Filter (KF) for linear estimation, or the Extended Kalman Filter (EKF) for use on non-linear problems. However certain assumptions made in the derivation of the EKF algorithms render it sub-optimal for aerial obstacle track estimation. Additionally the EKF has problems with initialisation and divergence (stability) for many non-linear processes. Research at the University of Southampton has highlighted a link between fuzzy networks and associative memory neural networks. This link is important as it allows new learning rules to be developed for training fuzzy rules, and learning convergence to be proved. This paper will explore methods for the fusion of estimates using these neurofuzzy models, and also address some of the weakness of the Kalman filter approximation introduced by the assumptions made in its derivation. 1996 Doyle R.S. Harris C.J. confpaper Modelling and Control of Nonlinear, Operating Point dependent Systems via Associative Memory Networks 2 This paper presents a novel approach to the modelling and control of a specific class of nonlinear systems whose parameters are unknown nonlinear functions of the measurable operating points. An associative memory network is used to identify each nonlinear function, whose inputs are the measurable operating points and output being the estimated value of the parameter. Two different cases are considered; the first being those systems where the networks can exactly model the nonlinear functions, whereas the second case considers those systems which can only approximate the nonlinear functions to a known accuracy. The first type of system is referred to as a matching system and the second is called a mismatching system. During the modelling phase, the weights for each network are trained in parallel using the normalised back-propagation algorithm for matching systems, and the modified recursive least squares algorithm for mismatching systems. It has been shown that these algorithms, together with Goodwin's technical lemma lead to a stable d-step-ahead control scheme for matching systems and a pole assignment control strategy for mismatching systems. 1996 Wang H. Brown M. Harris C.J. confpaper Vehicle Detection and Recognition in Greyscale Imagery 4 This paper details a novel two-stage vehicle detection and recognition algorithm by combining an image-processing region of interest (ROI) designator to cue a secondary recognition process implemented using principal component analysis (PCA) as input to a Multi-Layered Perceptron (MLP) classifier. Both the image-processing system and the MLP classifier have been designed for real-time implementation and data-fusion with other information sources. 1996 Matthews N.D. An P.E. Charnley D. Harris C.J. confpaper Understanding chaotic dissipative dynamics in the State Space with Fuzzy Systems The introduction gives an overview of chaotic dynamics and their particular properties. The information theoretical approach of unbiased guess, which was proposed by Jaynes, is utilized to derive a probability distribution in the state space. The weighting of parts in the state space by fuzzy sets provides additional information which enables a ``reconstruction'' of probabilities of crisp elements in state spaces without explicitely given crisp boxes and their attached probabilities, when the resolution of the fuzzy sets is fine enough. After summerizing some of the probabilistic quantities for the qualitative description of chaotic maps, a simple example of two fuzzy sets bounded by a crisp interval is employed to demonstrate by comparing the numerical results together with an analytical map, how this approach may be used to calculate probability densities, which are a basic quantity for chaos understanding. 1996 Schilhabel T.E. Harris C.J. confpaper Mission Management System for Multiple Autonomous Vehicles John Wiley and Sons 1996 Rayner N.J.W Harris C.J Baldwin J.F. confpaper Neurofuzzy Algorithms for Model Identification: Structure and Parameter Determination This paper describes some of the issues associated with developing a class of neurofuzzy construction algorithms based on B-spline fuzzy membership functions. These techniques have many desirable properties and links can be made with more conventional statistical model building approaches. The neurofuzzy model is decomposed into its linear and nonlinear components and a search technique is used to identify the structural nonlinearities whereas standard linear optimisation algorithms are used to identify the linear parameters. This paper discusses the performance of these two elements and contrasts their roles in the context of neurofuzzy systems. 1996 Brown M. Bossley K.M. Harris C.J. confpaper Intelligent Modelling, Control and Navigation for {AUV}s The demand for long mission duration, associated reliability and availability requirements, operation in unbinign, unstructured environments, real time operating requirements, and flexibility of use has led to increasing levels of autonomy in underwater vehicles. Autonomy in above water vehicles is quite old, with modern cruise missiles exemplifying extreme performance capability in navigation and control in complex environments under sever counter measure attack! Remotely piloted (air) vehicles for surveillance for civil and military usage are both sophisticated and common place today, but offer little intelligence and autonomy in that the majority are teleoperated or at most under significant levels of supervised control/management, with the majority of processing being carried out at the command and control centre. Whilst this approach minimisers the cost of onboard processing, it demands very high communication bandwidth, and lack of flexibility and robustness. It has been demonstrated that decentralised-distributed systems architectures potentially offer a wide range of quality attributes such as ease of systems integration, interoperability, scaling, portability and modularity, inherent robustness and survivability etc, that are necessary for effective implementation of AUVs. Such an architecture must be applicable to all the subsystems necessary for constructing AUVs ie. vehicle locatistaion, obstacle detection, vehicle navigation (and guidance), vehicle control, vehicle local planning and replanning and resource management, and vehicle mission tasking. For the purposes of this talk we will focus on the problems associated with vehicle state determination and control - since they are critical to all other AUV requirements and are relatively well understood and researched. To achieve improved quality of estimation, robustness and fault tolerance, multiple disparate sensors are increasingly being utilised in AUVs, so that following a particular sensor failure other sensors can cover the loss and at worst provide graceful degrading system performance. This in turn requires a mechanism for fusing data from disparate data sources (including symbolic and linguistic) and associated estimation algorithms for dealing with uncertainty and ignorance of coverage etc. Aspects of multi-sensor data fusion for vehicle localisation and obstacle detection, together with developments in estimation algorithms will be covered in the presentation. The generation of full envelope control laws for AUVs that accommodate mass temperature, salinity, depth, etc changes that in turn minimises fuel usage and is robust to damage etc, is extremely difficult by conventional controller design methods. Usually, these methods require an exchange between performance and robustness. In this paper we illustrate AUV modelling and control which enables both criteria to be satisfied utilising Neurofuzzy methods and online data gathering via the sensor bed. The talk will be illustrated by a selection of videos illustrating land based intelligent autonomous vehicles. 1996 Harris C.J. confpaper Bayesian Regularisation Applied to Neurofuzzy models Due to the inherent structure of neurofuzzy systems they are prone to poor generalisation. This paper discusses two complementary methods which should be employed to maximally exploit the available linguistic and numerical data, to overcome this problem. The main emphasis is the application of Bayesian regularisation to additive B-spline neurofuzzy models. This produces models which generalise well with poor quality data and hence identify more reliable rule bases. Error bars are then used to identify possible weakness in the resulting rule base, which require further validation. 1996 Bossley K.M. Brown M. Harris C.J. confpaper The Development and Application of Neurofuzzy Systems IEE Fuzzy and neurofuzzy systems have been widely applied in the domestic products during the past 8 years for a number of reasons. In 1989, a large, 5 year fuzzy logic research and applications program was funded in Japan by MITI which involved over 45 industrial companies including most of the major automotive and electrical manufacturers. One of the main topics of this research was finding ways to make machines smarter by increasing the number of sensors and developing new ways to make use of this extra information. Sometimes the sensor only existed in algorithmic form (a soft sensor), where other measurements were combined to infer the value of a particular variable. Products which involved fuzzy systems therefore, generally either performed better or were easier to use, and as such were readily accepted by consumers in the Far East. Factors, such as a reduced product development times and a flexible framework, were cited as reasons why fuzzy techniques were a good solution, but these must be balanced against the fact that it would take just as long to validate the final solution and that the fuzzy solution was often the only system developed. The interest in fuzzy systems which occurred in Europe and America during the Nineties was partially in response to the wave of electrical consumer applications which appeared in the Far East. The first fuzzy control system was developed in the UK over 20 years ago, and during the Seventies, Mamdani and hi co-workers at Queen Mary College, London, developed a number of static and adaptive fuzzy PID-type controllers. This work involved testing their performance against standard PID controllers on a range of different plants. However, the first real applications of fuzzy control were developed in Holland in 1979 and was used to control a cement kiln. Peter Holmbald had attended a session where Mamdani presented a paper on self-organising controllers which concluded at the end that {\em self-organising controllers were complicated and that he had recently stumbled on a better method: fuzzy logic}. Fuzzy logic was therefore proposed as a method for simplifying the controller design process, making it: quicker to design, perform better than standard linear methods and less reliant on an accurate mathematical model of the plan