Skip to main content

M. Rajeswari

Universiti Sains Malaysia, School of Computer Sciences, Faculty Member

Followers

5

Following

3

Co-authors

2

Public Views

Graduate Center of the City University of New York

Marko Ala-Fossi

Tampere University

University of Lucerne

Guido Nockemann

Friedrich-Alexander-Universität Erlangen-Nürnberg

Mehmet Fatih Amasyali

Yildiz Technical University

Swansea University

Birzeit University

Historic Environment Scotland

German Research Center for Artificial Intelligence

Punjab Technical University

InterestsView All (6)

Uploads

Papers by M. Rajeswari

Automatic Overlaid Text Detection, Extraction and Recognition for High Level Event/Concept Identification in Soccer Videos

2008 International Conference on Computer and Electrical Engineering, 2008

Shot view classification for playfield-based sports video

2009 IEEE International Conference on Signal and Image Processing Applications, 2009

Overlaid Text Recognition for Matching Soccer-Concept Keywords

2008 Fifth International Conference on Computer Graphics, Imaging and Visualisation, 2008

Goal Event Detection in Soccer Videos via Collaborative Multimodal Analysis

Detecting semantic events in sports video is crucial for video indexing and retrieval. Most exist... more Detecting semantic events in sports video is crucial for video indexing and retrieval. Most existing works have exclusively relied on video content features, namely, directly available and extractable data from the visual and/or aural channels. Sole reliance on such data however, can be problematic due to the high-level semantic nature of video and the difficulty to properly align detected events with their exact time of occurrences. This paper proposes a framework for soccer goal event detection through collaborative analysis of multimodal features. Unlike previous approaches, the visual and aural contents are not directly scrutinized. Instead, an external textual source (i.e., minute-by-minute reports from sports websites) is used to initially localize the event search space. This step is vital as the event search space can significantly be reduced. This also makes further visual and aural analysis more efficient since excessive and unnecessary non-eventful segments are discarded, culminating in the accurate identification of the actual goal event segment. Experiments conducted on thirteen soccer matches are very promising with high accuracy rates being reported.

Soccer Event Detection via Collaborative Multimodal Feature Analysis and Candidate Ranking

Multimodal Integration (Image and Text) Using Ontology Alignment

American Journal of Applied Sciences, 2009

Automatic Image Annotation using Mixtures of the Exponential Family

Journal of Convergence Information Technology, 2011

Image annotation is an important task in computer vision. The annotated images can be very useful... more Image annotation is an important task in computer vision. The annotated images can be very useful for indexing and retrieval applications. In this paper, we propose a generative model for image annotation based on mixtures of the exponential family of distributions. The distributions considered are the Multivariate Gaussian, Rayleigh, Poisson, Bernoulli and Centered Laplacian. The model leads to a generic algorithm that is able to perform learning in a more efficient and flexible manner. Subsequently, it can be used to evaluate the performance of each distribution in the task of image annotation. The approach is also compared to a discriminative approach (i.e. Support Vector Machine) using the LabelMe dataset for the concepts Buildings, Street, Mountains and Coasts, where promising classification results were reported for some of the concepts.

A Review on Ontology-Based Label Extraction from Image Data

Ontology-based label extraction is extensively used to interpret the semantics found in image and... more Ontology-based label extraction is extensively used to interpret the semantics found in image and video data. Particularly, ontology-based label extraction is one of the main steps in object class recognition, image annotation, and image disambiguation. These applications have important roles in the field of image analysis, and as such, a number of variations of the ontology-based label extraction used in these applications have been reported in the literature. These variations involve ontology development and utilization, and can affect the applicability (e.g., domain-and application-dependency) as well as the accuracy of the output. Unfortunately, the variability aspect of this variation has neither been established nor tracked. Thus, the variations were not configured. A review of the ontology-based label extraction based on the input data, the utilized technique, and the type of utilized ontology is presented in this paper. The ontology-based label extraction is categorized based on two aspects, namely, the type of input data and the type of ontology used. These two aspects determine the type of the label extraction technique to be used. As a result, the relative advantages and disadvantages of each category are determined. The gaps and future research directions in this field are also highlighted.

Similarity measure determination from NURBS-warping method

7th International Conference on Control, Automation, Robotics and Vision, 2002. ICARCV 2002., 2002

With the advances in digital imagery, large accessible data storage, internet repositories, and i... more With the advances in digital imagery, large accessible data storage, internet repositories, and image applications, information conveyed through images is gaining in importance. There is a need to find a desired image from a collection of images, which is shared by many groups including journalists, engineers, historians, designers, teachers, artists and advertising agencies. Humans have the capability to determine only a few selected signs, which enable them to determine the impression of a complete representation of the objects and comparing them. However, it is impossible to trace through the large image database in order to retrieve the desired image manually. Retrieval by shape is considered one of the most difficult and challenging aspects of image-based search. Thus, an effective shape retrieval system approach, which resembles the human capability in retrieving images from a database that are relevant to a query, is needed. In this paper, we propose a new similarity measure determination based on the NURBS-Warping method. In this method, the query image represented by the Non-Uniform Rational B-Spline (NURBS) descriptor is warped to best fit the shape of the database image. The similarity between two images is determined by measuring the effort been spent in the warping process. Four similarity parameters are derived to reflect the effort spent throughout the warping process. These parameters are the Total Cumulative Change in the Position of the Control Points (TCCPCP), Total Cumulative Change in the Curvature of the Control Points (TCCCCP), Total Cumulative Voting Distance Score (TCVDS) and degree of matching, which are used to compute the similarity measure for each database image. The retrieved database images are ranked according to the computed similarity measure. A set of experiments is carried out to investigate the characteristic of each similarity parameters that influence the similarity measure computation independently. We also validate the effectiveness of the computed similarity measure for retrieving a set of database images based on the human similarity judgement. The retrieval results of the proposed method are compared with the Curvature Scale Space (CSS) method proposed by Mokhtarian. The overall experiments results show that the similarity measure derived from the NURBS-Warping method show good retrieval results in a highly similar database.

$Research paper thumbnail of A fractal modeling for network traffic$

A fractal modeling for network traffic

2000 TENCON Proceedings. Intelligent Systems and Technologies for the New Millennium (Cat. No.00CH37119), 2000

This paper describes fractal modeling with application to broadcast traflc of switch based LAN. T... more This paper describes fractal modeling with application to broadcast traflc of switch based LAN. The traflc is represented by two variables, packet interarrival time and packet size. The series of these 2 variables is normalized separately and considered as a number of random points in 2-dimensional space. In the stu& fractal behavior of the traflc is considered to be caused by high variability of packet inter-arrival time. Hence measuring variability of inter-arrival time is a necessity for fractal analysis. This analysis is done using long-range dependence (LRD) of variance measurement. The work compares LRD properties between real broadcast traflc and the synthetic one. The real traflc is captured from switch based-LAN, while the synthetic one is generated using Iterated Function System (IFS) method

Using Correlation Based Subspace Clustering for Multi-label Text Data Classification

2010 22nd IEEE International Conference on Tools with Artificial Intelligence, 2010

AbstractWith the boom of web and social networking, the amount of generated text data has increa... more

Wavelet Energy Guided Level Set for Segmenting Highly Similar Regions in Medical Images

2009 2nd International Conference on Biomedical Engineering and Informatics, 2009

Abstract This paper presents a new segmentation method that integrates a wavelet based feature, ... more

Segmentation of Abdominal Volume Dataset Slices Guided by Single Annotated Image

2009 2nd International Conference on Biomedical Engineering and Informatics, 2009

One major challenge faced by segmentation techniques in analyzing and visualizing individual slic... more One major challenge faced by segmentation techniques in analyzing and visualizing individual slices of a 3D anatomical structure, is the degree of manual interaction required. To alleviate this problem, researchers have proposed the automatic incorporation of anatomical knowledge, via medical atlases to assist with the segmentation process. Some solutions include constructing specialized simple 2D, as well as complex 3D atlases. In this paper we propose a simple method that automatically transfers a prior anatomical knowledge from a simple teaching atlas of a single 2D slice to the most similar slice in the 3D volume dataset. Segmentation of the selected anatomy is then able to be propagated automatically on the remaining slices in the dataset without further manual interaction. Our experiments are conducted on abdomen CT images and we successful delineation and visualization have been demonstrated for spleen, and kidney images.

Image segmentation with cyclic load balanced parallel Fuzzy C - Means cluster analysis

2011 IEEE International Conference on Imaging Systems and Techniques, 2011

... Parallel Fuzzy C – Means Cluster Analysis Mogana Vadiveloo, Rosni Abdullah, Mandava Rajeswari... more

Segmentation And Quantification Of The Cupriavidus Sp. Bacterium Using Microscopy Images

2008 International Conference on Signal Processing, Communications and Networking, 2008

... Page 4. Segmentation And Quantification OfThe Cupriavidus Sp. Bacterium Using Microscopy Imag... more

Real-time Analysis of an IC Wire-bonding Inspection System

Real-Time Imaging, 1999

Real-Time 3D Motion Tracking with Known Geometric Models n this paper a new model-based tracking ... more Real-Time 3D Motion Tracking with Known Geometric Models n this paper a new model-based tracking algorithm is proposed for real-time performance. The matching process includes two aspects of: (1) feature extraction using local minimum energy and I(2) global matching of known 3D models against the projected features. The algorithm is robust to change in lighting and background. The small motion hypothesis was used for fitting of feature energy which is defined as the negative absolute value of the edge strength. An autoregressive AR(1) model is employed for detecting incorrect matches in terms of the feature energy. We have found a new invariance-based method to eliminate false matches caused by strong shadow or occlusion. The invariance is the ratio of trigonometric functions of the angles formed by a polygon. In order to calculate the vertices of the object surface in an image, regression technique in terms of matched features is efficient in our approach. A linear least squares method and the orthonormal rotation matrix are used for motion estimation and pose update of the six degrees of freedom. Also, an Extended Kalman Filter is introduced to guarantee a smooth motion estimation and prediction.

Multiobjective clustering with metaheuristic: current trends and methods in image segmentation

IET Image Processing, 2012

We develop a metaheuristic procedure for multiobjective clustering problems. Our goal is to find ... more We develop a metaheuristic procedure for multiobjective clustering problems. Our goal is to find good approximations of the efficient frontier for this class of problems and provide a means for improving decision making in multiple areas of application and in particular those related to marketing. The procedure is based on the tabu and scatter search methodologies. Clustering problems have been the subject of numerous studies; however, most of the work has focused on single-objective problems. Clustering using multiple criteria and/or multiple data sources has received limited attention in the OR and marketing literature. Our procedure is general and tackles several problems classes within this area of combinatorial data analysis. We conduct extensive experimentation with both artificial and real data (in a marketing-segmentation problem) to show the effectiveness of the proposed procedure.

Novel Direct and Self-Regulating Approaches to Determine Optimum Growing Multi-Experts Network Structure

IEEE Transactions on Neural Networks, 2004

This paper presents two novel approaches to determine optimum growing multi-experts network (GMN)... more This paper presents two novel approaches to determine optimum growing multi-experts network (GMN) structure. The first method called direct method deals with expertise domain and levels in connection with local experts. The growing neural gas (GNG) algorithm is used to cluster the local experts. The concept of error distribution is used to apportion error among the local experts. After reaching the specified size of the network, redundant experts removal algorithm is invoked to prune the size of the network based on the ranking of the experts. However, GMN is not ergonomic due to too many network control parameters. Therefore, a self-regulating GMN (SGMN) algorithm is proposed. SGMN adopts self-adaptive learning rates for gradient-descent learning rules. In addition, SGMN adopts a more rigorous clustering method called fully self-organized simplified adaptive resonance theory in a modified form. Experimental results show SGMN obtains comparative or even better performance than GMN in four benchmark examples, with reduced sensitivity to learning parameters setting. Moreover, both GMN and SGMN outperform the other neural networks and statistical models. The efficacy of SGMN is further justified in three industrial applications and a control problem. It provides consistent results besides holding out a profound potential and promise for building a novel type of nonlinear model consisting of several local linear models. Index Terms-Growing neural network modular neural network, local linear model.

Growing multi-experts network with confidence interval estimation

Extrapolation detection and novelty-based node insertion for sequential growing multi-experts network

Applied Soft Computing, 2003

Artificial neural networks (ANNs) have been used to construct empirical nonlinear models of proce... more Artificial neural networks (ANNs) have been used to construct empirical nonlinear models of process data. Because networks are not based on physical theory and contain nonlinearities, their predictions are suspect when extrapolating beyond the range of original training data. Standard networks give no indication of possible errors due to extrapolation. This paper describes a sequential supervised learning scheme for the recently formalized Growing multi-experts network (GMN). It is shown that certainty factor can be generated by GMN that can be taken as extrapolation detector for GMN. On-line GMN identification algorithm is presented and its performance is evaluated. The capability of the GMN to extrapolate is also indicated. Four benchmark experiments are dealt with to demonstrate the effectiveness and utility of GMN as a universal function approximator.

Automatic Overlaid Text Detection, Extraction and Recognition for High Level Event/Concept Identification in Soccer Videos

2008 International Conference on Computer and Electrical Engineering, 2008

Shot view classification for playfield-based sports video

2009 IEEE International Conference on Signal and Image Processing Applications, 2009

Overlaid Text Recognition for Matching Soccer-Concept Keywords

2008 Fifth International Conference on Computer Graphics, Imaging and Visualisation, 2008

Goal Event Detection in Soccer Videos via Collaborative Multimodal Analysis

Detecting semantic events in sports video is crucial for video indexing and retrieval. Most exist... more Detecting semantic events in sports video is crucial for video indexing and retrieval. Most existing works have exclusively relied on video content features, namely, directly available and extractable data from the visual and/or aural channels. Sole reliance on such data however, can be problematic due to the high-level semantic nature of video and the difficulty to properly align detected events with their exact time of occurrences. This paper proposes a framework for soccer goal event detection through collaborative analysis of multimodal features. Unlike previous approaches, the visual and aural contents are not directly scrutinized. Instead, an external textual source (i.e., minute-by-minute reports from sports websites) is used to initially localize the event search space. This step is vital as the event search space can significantly be reduced. This also makes further visual and aural analysis more efficient since excessive and unnecessary non-eventful segments are discarded, culminating in the accurate identification of the actual goal event segment. Experiments conducted on thirteen soccer matches are very promising with high accuracy rates being reported.

Soccer Event Detection via Collaborative Multimodal Feature Analysis and Candidate Ranking

Multimodal Integration (Image and Text) Using Ontology Alignment

American Journal of Applied Sciences, 2009

Automatic Image Annotation using Mixtures of the Exponential Family

Journal of Convergence Information Technology, 2011

Image annotation is an important task in computer vision. The annotated images can be very useful... more Image annotation is an important task in computer vision. The annotated images can be very useful for indexing and retrieval applications. In this paper, we propose a generative model for image annotation based on mixtures of the exponential family of distributions. The distributions considered are the Multivariate Gaussian, Rayleigh, Poisson, Bernoulli and Centered Laplacian. The model leads to a generic algorithm that is able to perform learning in a more efficient and flexible manner. Subsequently, it can be used to evaluate the performance of each distribution in the task of image annotation. The approach is also compared to a discriminative approach (i.e. Support Vector Machine) using the LabelMe dataset for the concepts Buildings, Street, Mountains and Coasts, where promising classification results were reported for some of the concepts.

A Review on Ontology-Based Label Extraction from Image Data

Ontology-based label extraction is extensively used to interpret the semantics found in image and... more Ontology-based label extraction is extensively used to interpret the semantics found in image and video data. Particularly, ontology-based label extraction is one of the main steps in object class recognition, image annotation, and image disambiguation. These applications have important roles in the field of image analysis, and as such, a number of variations of the ontology-based label extraction used in these applications have been reported in the literature. These variations involve ontology development and utilization, and can affect the applicability (e.g., domain-and application-dependency) as well as the accuracy of the output. Unfortunately, the variability aspect of this variation has neither been established nor tracked. Thus, the variations were not configured. A review of the ontology-based label extraction based on the input data, the utilized technique, and the type of utilized ontology is presented in this paper. The ontology-based label extraction is categorized based on two aspects, namely, the type of input data and the type of ontology used. These two aspects determine the type of the label extraction technique to be used. As a result, the relative advantages and disadvantages of each category are determined. The gaps and future research directions in this field are also highlighted.

Similarity measure determination from NURBS-warping method

7th International Conference on Control, Automation, Robotics and Vision, 2002. ICARCV 2002., 2002

With the advances in digital imagery, large accessible data storage, internet repositories, and i... more With the advances in digital imagery, large accessible data storage, internet repositories, and image applications, information conveyed through images is gaining in importance. There is a need to find a desired image from a collection of images, which is shared by many groups including journalists, engineers, historians, designers, teachers, artists and advertising agencies. Humans have the capability to determine only a few selected signs, which enable them to determine the impression of a complete representation of the objects and comparing them. However, it is impossible to trace through the large image database in order to retrieve the desired image manually. Retrieval by shape is considered one of the most difficult and challenging aspects of image-based search. Thus, an effective shape retrieval system approach, which resembles the human capability in retrieving images from a database that are relevant to a query, is needed. In this paper, we propose a new similarity measure determination based on the NURBS-Warping method. In this method, the query image represented by the Non-Uniform Rational B-Spline (NURBS) descriptor is warped to best fit the shape of the database image. The similarity between two images is determined by measuring the effort been spent in the warping process. Four similarity parameters are derived to reflect the effort spent throughout the warping process. These parameters are the Total Cumulative Change in the Position of the Control Points (TCCPCP), Total Cumulative Change in the Curvature of the Control Points (TCCCCP), Total Cumulative Voting Distance Score (TCVDS) and degree of matching, which are used to compute the similarity measure for each database image. The retrieved database images are ranked according to the computed similarity measure. A set of experiments is carried out to investigate the characteristic of each similarity parameters that influence the similarity measure computation independently. We also validate the effectiveness of the computed similarity measure for retrieving a set of database images based on the human similarity judgement. The retrieval results of the proposed method are compared with the Curvature Scale Space (CSS) method proposed by Mokhtarian. The overall experiments results show that the similarity measure derived from the NURBS-Warping method show good retrieval results in a highly similar database.

$Research paper thumbnail of A fractal modeling for network traffic$

A fractal modeling for network traffic

2000 TENCON Proceedings. Intelligent Systems and Technologies for the New Millennium (Cat. No.00CH37119), 2000

This paper describes fractal modeling with application to broadcast traflc of switch based LAN. T... more This paper describes fractal modeling with application to broadcast traflc of switch based LAN. The traflc is represented by two variables, packet interarrival time and packet size. The series of these 2 variables is normalized separately and considered as a number of random points in 2-dimensional space. In the stu& fractal behavior of the traflc is considered to be caused by high variability of packet inter-arrival time. Hence measuring variability of inter-arrival time is a necessity for fractal analysis. This analysis is done using long-range dependence (LRD) of variance measurement. The work compares LRD properties between real broadcast traflc and the synthetic one. The real traflc is captured from switch based-LAN, while the synthetic one is generated using Iterated Function System (IFS) method

Using Correlation Based Subspace Clustering for Multi-label Text Data Classification

2010 22nd IEEE International Conference on Tools with Artificial Intelligence, 2010

AbstractWith the boom of web and social networking, the amount of generated text data has increa... more

Wavelet Energy Guided Level Set for Segmenting Highly Similar Regions in Medical Images

2009 2nd International Conference on Biomedical Engineering and Informatics, 2009

Abstract This paper presents a new segmentation method that integrates a wavelet based feature, ... more

Segmentation of Abdominal Volume Dataset Slices Guided by Single Annotated Image

2009 2nd International Conference on Biomedical Engineering and Informatics, 2009

One major challenge faced by segmentation techniques in analyzing and visualizing individual slic... more One major challenge faced by segmentation techniques in analyzing and visualizing individual slices of a 3D anatomical structure, is the degree of manual interaction required. To alleviate this problem, researchers have proposed the automatic incorporation of anatomical knowledge, via medical atlases to assist with the segmentation process. Some solutions include constructing specialized simple 2D, as well as complex 3D atlases. In this paper we propose a simple method that automatically transfers a prior anatomical knowledge from a simple teaching atlas of a single 2D slice to the most similar slice in the 3D volume dataset. Segmentation of the selected anatomy is then able to be propagated automatically on the remaining slices in the dataset without further manual interaction. Our experiments are conducted on abdomen CT images and we successful delineation and visualization have been demonstrated for spleen, and kidney images.

Image segmentation with cyclic load balanced parallel Fuzzy C - Means cluster analysis

2011 IEEE International Conference on Imaging Systems and Techniques, 2011

... Parallel Fuzzy C – Means Cluster Analysis Mogana Vadiveloo, Rosni Abdullah, Mandava Rajeswari... more

Segmentation And Quantification Of The Cupriavidus Sp. Bacterium Using Microscopy Images

2008 International Conference on Signal Processing, Communications and Networking, 2008

... Page 4. Segmentation And Quantification OfThe Cupriavidus Sp. Bacterium Using Microscopy Imag... more

Real-time Analysis of an IC Wire-bonding Inspection System

Real-Time Imaging, 1999

Real-Time 3D Motion Tracking with Known Geometric Models n this paper a new model-based tracking ... more Real-Time 3D Motion Tracking with Known Geometric Models n this paper a new model-based tracking algorithm is proposed for real-time performance. The matching process includes two aspects of: (1) feature extraction using local minimum energy and I(2) global matching of known 3D models against the projected features. The algorithm is robust to change in lighting and background. The small motion hypothesis was used for fitting of feature energy which is defined as the negative absolute value of the edge strength. An autoregressive AR(1) model is employed for detecting incorrect matches in terms of the feature energy. We have found a new invariance-based method to eliminate false matches caused by strong shadow or occlusion. The invariance is the ratio of trigonometric functions of the angles formed by a polygon. In order to calculate the vertices of the object surface in an image, regression technique in terms of matched features is efficient in our approach. A linear least squares method and the orthonormal rotation matrix are used for motion estimation and pose update of the six degrees of freedom. Also, an Extended Kalman Filter is introduced to guarantee a smooth motion estimation and prediction.

Multiobjective clustering with metaheuristic: current trends and methods in image segmentation

IET Image Processing, 2012

We develop a metaheuristic procedure for multiobjective clustering problems. Our goal is to find ... more We develop a metaheuristic procedure for multiobjective clustering problems. Our goal is to find good approximations of the efficient frontier for this class of problems and provide a means for improving decision making in multiple areas of application and in particular those related to marketing. The procedure is based on the tabu and scatter search methodologies. Clustering problems have been the subject of numerous studies; however, most of the work has focused on single-objective problems. Clustering using multiple criteria and/or multiple data sources has received limited attention in the OR and marketing literature. Our procedure is general and tackles several problems classes within this area of combinatorial data analysis. We conduct extensive experimentation with both artificial and real data (in a marketing-segmentation problem) to show the effectiveness of the proposed procedure.

Novel Direct and Self-Regulating Approaches to Determine Optimum Growing Multi-Experts Network Structure

IEEE Transactions on Neural Networks, 2004

This paper presents two novel approaches to determine optimum growing multi-experts network (GMN)... more This paper presents two novel approaches to determine optimum growing multi-experts network (GMN) structure. The first method called direct method deals with expertise domain and levels in connection with local experts. The growing neural gas (GNG) algorithm is used to cluster the local experts. The concept of error distribution is used to apportion error among the local experts. After reaching the specified size of the network, redundant experts removal algorithm is invoked to prune the size of the network based on the ranking of the experts. However, GMN is not ergonomic due to too many network control parameters. Therefore, a self-regulating GMN (SGMN) algorithm is proposed. SGMN adopts self-adaptive learning rates for gradient-descent learning rules. In addition, SGMN adopts a more rigorous clustering method called fully self-organized simplified adaptive resonance theory in a modified form. Experimental results show SGMN obtains comparative or even better performance than GMN in four benchmark examples, with reduced sensitivity to learning parameters setting. Moreover, both GMN and SGMN outperform the other neural networks and statistical models. The efficacy of SGMN is further justified in three industrial applications and a control problem. It provides consistent results besides holding out a profound potential and promise for building a novel type of nonlinear model consisting of several local linear models. Index Terms-Growing neural network modular neural network, local linear model.

Growing multi-experts network with confidence interval estimation

Extrapolation detection and novelty-based node insertion for sequential growing multi-experts network

Applied Soft Computing, 2003

Artificial neural networks (ANNs) have been used to construct empirical nonlinear models of proce... more Artificial neural networks (ANNs) have been used to construct empirical nonlinear models of process data. Because networks are not based on physical theory and contain nonlinearities, their predictions are suspect when extrapolating beyond the range of original training data. Standard networks give no indication of possible errors due to extrapolation. This paper describes a sequential supervised learning scheme for the recently formalized Growing multi-experts network (GMN). It is shown that certainty factor can be generated by GMN that can be taken as extrapolation detector for GMN. On-line GMN identification algorithm is presented and its performance is evaluated. The capability of the GMN to extrapolate is also indicated. Four benchmark experiments are dealt with to demonstrate the effectiveness and utility of GMN as a universal function approximator.