Search

The problem of subspace clustering which refers to segmenting a collection of data samples approximately drawn from a union of linear subspaces is considered in this paper. Among existing subspace clustering algorithms, low rank representation (LRR) based subspace clustering is a very powerful method and has demonstrated that its performance is good. Latent low rank representation (LLRR) subspace clustering algorithm is an improvement of the original LRR algorithm when the observed data samples are insufficient. The clustering accuracy of LLRR is higher than that of LRR. Recently, Frobenius norm minimization based LRR algorithm has been proposed and its clustering accuracy is higher than that of LRR demonstrating the effectiveness of Frobenius norm as another convex surrogate of the rank function. Combining LLRR and Frobenius norm, a new low rank representation subspace clustering algorithm is proposed in this paper. The nuclear norm in the LLRR algorithm is replaced by the Frobenius norm. The resulting optimization problem is solved via alternating direction method of multipliers (ADMM). Experimental results show that compared with LRR, LLRR and several other state-of-the-art subspace clustering algorithms, the proposed algorithm can get higher clustering accuracy. Compared with LLRR, the running time of the proposed algorithm is reduced significantly. [ABSTRACT FROM AUTHOR]

This paper mainly focuses on how to construct concept lattice effectively and efficiently based on improved variable precision rough set. On the basis of preprocessing formal concept, one algorithm that can determine the value range of variable precision parameter β according to the approximate classification quality is proposed. An improved β -upper and lower distribution attribute reduction algorithm is also proposed based on the improved variable precision rough set, the algorithm can be used for attribute reduction on the original data of the concept lattice, and to eliminate the redundant knowledge or noises of the formal context. For the reduced formal context, the paper combines the concept construction algorithm with an improved rule acquisition algorithm seamlessly, and proposes a novel approach of concept lattice construction based on improved variable precision rough set. Finally, a concept lattice generation prototype system is developed, this paper also performs comprehensive experiments, and the effectiveness of the improved algorithm is proved through the experimental results. [ABSTRACT FROM AUTHOR]

This paper proposes an efficient and effective active online sequential learning approach, named as Less Annotated Active Learning Extreme Learning Machine (LAAL-ELM). It leverages the predictions’ confidence of the new arriving data to actively select both query-annotated samples and confidence-weighted predict-annotated ones to update the classifier, which contributes to less actively query annotation, and applies WOS-ELM, a discriminant model, to significantly reduce the computation complexity for doing online updating in one step. The proposed approach firstly gives a principle to evaluate confidence of the prediction in WOS-ELM; then determines what and how to update the model with new arriving data in the online phase: the uncertain instances are annotated by query their classes, almost-certain ones are weighted on its prediction’s confidence and the certain ones are discarded directly for reducing over-fitting; at last, the weighted and query-annotated samples are used to update the classifier. The proposed approach is evaluated on five real-world benchmark classification issues. And the experimental results demonstrate that the proposed LAAL-ELM can effectively reduce the number of queried samples while maintaining high level of classification performance. [ABSTRACT FROM AUTHOR]

This paper presents a novel unsupervised multi-source domain adaptation approach, named as coupled local–global adaptation (CLGA). At the global level, in order to maximize the adaptation ability, CLGA regards multiple domains as a unity, and jointly mitigates the gaps of both marginal and conditional distributions between source and target dataset. At the local level, with the intention of maximizing the discriminative ability, CLGA investigates the relationship among distinctive domains, and exploits both class and domain manifold structures embedded in data samples. We formulate both local and global adaptation in a concise optimization problem, and further derive an analytic solution for the objective function. Extensive evaluations verify that CLGA performs better than several existing methods not only in multi-source adaptation tasks but also in single source scenarios. [ABSTRACT FROM AUTHOR]

Compressed sensing (CS) has been one of the great successes of applied mathematics in the last decade. This paper proposes a new method, combining the advantage of the Compressive Sampling Matching Pursuit (CoSaMP) algorithm and the Quasi–Newton Iteration Projection (QNIP) algorithm, for the recovery of sparse signal from underdetermined linear systems. To get the new algorithm, Quasi–Newton Projection Pursuit (QNPP), the least-squares technique in CoSaMP is used to accelerate convergence speed and QNIP is modified slightly. The convergence rate of QNPP is studied, under a certain condition on the restricted isometry constant of the measurement matrix, which is smaller than that of QNIP. The fast version of QNPP is also proposed which uses the Richardson’s iteration to reduce computation time. The numerical results show that the proposed algorithms have higher recovery rate and faster convergence speed than existing techniques. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, a special class of recurrent neural network, termed Zhang neural network (ZNN), is investigated for the online solution of the time-varying matrix pseudoinverse. Meanwhile, a novel activation function, named Li activation function, is employed. Then, based on two basic Zhang functions (ZFs) and the intrinsically nonlinear method of ZNN design, two finite-time convergent ZNN models (termed ZNN-1 model and ZNN-2 model) are first proposed and investigated for time-varying matrix pseudoinversion. Such two ZNN models can be accelerated to finite-time convergence to the time-varying theoretical pseudoinverse. The upper bound of the convergence time is also derived analytically via Lyapunov theory. By exploiting the other three simplified ZFs and the extended nonlinearization method, three simplified finite-time convergent ZNN models (termed ZNN-3 model, ZNN-4 model and ZNN-5 model) are sequentially proposed. In addition, the link between the ZNN models and the Getz–Marsden (G–M) dynamic system is discovered and presented in this paper. Computer-simulation results further substantiate the theoretical analysis and demonstrate the effectiveness of ZNN models based on different ZFs for the time-varying matrix pseudoinverse. [Copyright &y& Elsevier]

Abstract: This paper introduces a new model for robot behavior categorization. Correlation based adaptive resonance theory (CobART) networks are integrated hierarchically in order to develop an adequate categorization, and to elicit various behaviors performed by the robot. The proposed model is developed by adding a second layer CobART network which receives first layer CobART network categories as an input, and back-propagates the matching information to the first layer networks. The first layer CobART networks categorize self-behavior data of a robot or an object in the environment while the second layer CobART network categorizes the robot''s behavior with respect to its effect on the object. Experiments show that the proposed model generates reasonable categorization of behaviors being tested. Moreover, it can learn different forms of the behaviors, and it can detect the relations between them. In essence, the model has an expandable architecture and it contains reusable parts. The first layer CobART networks can be integrated with other CobART networks for another categorization task. Hence, the model presents a way to reveal all behaviors performed by the robot at the same time. [Copyright &y& Elsevier]

Abstract: The objective of this paper is to present an efficient hardware architecture for generalized Hebbian algorithm (GHA). In the architecture, the principal component computation and weight vector updating of the GHA are operated in parallel, so that the throughput of the circuit can be significantly enhanced. In addition, the weight vector updating process is separated into a number of stages for lowering area costs and increasing computational speed. To show the effectiveness of the circuit, a texture classification system based on the proposed architecture is designed. It is embedded in a system-on-programmable-chip (SOPC) platform for physical performance measurement. Experimental results show that the proposed architecture is an efficient design for attaining both high speed performance and low area costs. [Copyright &y& Elsevier]

Abstract: This paper presents a new hybrid genetic algorithm (HGA) for feature selection (FS), called as HGAFS. The vital aspect of this algorithm is the selection of salient feature subset within a reduced size. HGAFS incorporates a new local search operation that is devised and embedded in HGA to fine-tune the search in FS process. The local search technique works on basis of the distinct and informative nature of input features that is computed by their correlation information. The aim is to guide the search process so that the newly generated offsprings can be adjusted by the less correlated (distinct) features consisting of general and special characteristics of a given dataset. Thus, the proposed HGAFS receives the reduced redundancy of information among the selected features. On the other hand, HGAFS emphasizes on selecting a subset of salient features with reduced number using a subset size determination scheme. We have tested our HGAFS on 11 real-world classification datasets having dimensions varying from 8 to 7129. The performances of HGAFS have been compared with the results of other existing ten well-known FS algorithms. It is found that, HGAFS produces consistently better performances on selecting the subsets of salient features with resulting better classification accuracies. [Copyright &y& Elsevier]

Abstract: In this paper, a new artificial neural network model is proposed for visual object recognition, in which the bottom-up, sensory-driven pathway and top-down, expectation-driven pathway are fused in information processing and their corresponding weights are learned based on the fused neuron activities. During the supervised learning process, the target labels are applied to update the bottom-up synaptic weights of the neural network. Meanwhile, the hypotheses generated by the bottom-up pathway produce expectations on sensory inputs through the top-down pathway. The expectations are constrained by the real data from the sensory inputs, which can be used to update the top-down synaptic weights accordingly. To further improve the visual object recognition performance, the multi-scale histograms of oriented gradients (MS-HOG) method is proposed to extract local features of visual objects from images. Extensive experiments on different image datasets demonstrate the efficiency and robustness of the proposed neural network model with features extracted using the MS-HOG method on visual object recognition compared with other state-of-the-art methods. [Copyright &y& Elsevier]

Abstract: The application of neural networks in the data mining has become wider. Although neural networks may have complex structure, long training time, and the representation of results is not comprehensible, neural networks have high acceptance ability for noisy data, high accuracy and are preferable in data mining. On the other hand, It is an open question as to what is the best way to train and extract symbolic rules from trained neural networks in domains like classification. In this paper, we train the neural networks by constructive learning and present the analysis of the convergence rate of the error in a neural network with and without threshold which have been learnt by a constructive method to obtain the simple structure of the network. The response of ANN is acquired but its result is not in understandable form or in a black box form. It is frequently desirable to use the model backwards and identify sets of input variable which results in a desired output value. The large numbers of variables and nonlinear nature of many materials models that can help finding an optimal set of difficult input variables. We will use a genetic algorithm to solve this problem. The method is evaluated on different public-domain data sets with the aim of testing the predictive ability of the method and compared with standard classifiers, results showed comparatively high accuracy. [Copyright &y& Elsevier]

Abstract: Some linear sparse coding models have been proposed for modeling responses in the early stage of the visual system, but nonlinear operations are ubiquitous in visual cortex. So we put forward an associative sparse coding neural network (ASCNN) with nonlinear response property in top-layer coding units. In our ASCNN model, the choices of sparseness function must correspond to the characters of activation function. This paper gives several reasonable and efficient methods for constructing sparseness functions and activation functions. Experiment on benchmark natural image dataset shows that our model can successfully simulate receptive field and nonlinear sparse response property of simple cells. Moreover, in two recognition tasks on face images and handwritten digits, experimental results show that our model works much better than linear sparse coding model (Sparsenet) by combining with linear neural network classifier. [Copyright &y& Elsevier]

Abstract: In this paper, the neural network composed of a chaotic neuron layer and a linear neuron layer is used to construct a block cipher that transforms the data from the plaintext form into the unintelligible form under the control of the key. Among them, the chaotic neuron layer realizes data diffusion, the linear neuron layer realizes data confusion, and the two layers are repeated for several times to strengthen the cipher. The decryption process is symmetric to the encryption process. Theoretical analysis and experimental results show that the block cipher has good computing security and is more suitable for image encryption. It is expected to attract more researchers in this field. [Copyright &y& Elsevier]

Abstract: This paper studies the delay-interval-dependent stability of the equilibrium point of a general class of recurrent neural networks with time-varying delays that may exclude zero. By constructing the appropriate Lyapunov–Krasovskii functional, two sufficient conditions ensuring the global asymptotic stability of the equilibrium point of such networks with interval-time-varying delays are established. The present results, together with two numerical examples, show that the equilibrium points of the considered networks may be globally asymptotically stable in some delay interval(s) even though the equilibrium points of the corresponding delay-free recurrent neural networks are not globally asymptotically stable. [Copyright &y& Elsevier]

Abstract: In this paper, a new thresholding function is proposed for image denoising in the wavelet domain. The proposed function is further used in a new subband-adaptive thresholding neural network to improve the efficiency of the denoising procedure. Some new adaptive learning types are also proposed. In these learning methods, the threshold and the thresholding function effects are considered simultaneously. These methods are used to suppress two types of important noises, Gaussian and speckle, ranging from natural images to ultrasound and SAR pictures. The simulation results show that the proposed thresholding function has superior features compared to conventional methods when used with the proposed adaptive learning types. This makes it an efficient method in image denoising applications. [Copyright &y& Elsevier]

Abstract: Robust exponential stability for interval Cohen–Grossberg neural networks with time varying delay has received increasing interest in recent years. In this paper, some new criteria are derived using linear matrix inequality, matrix norm and Halanay inequality techniques. Compared with the existing results, these new criteria are not conservative and are convenient to check. Three numerical examples are used to show the effectiveness of the obtained results. [Copyright &y& Elsevier]

Abstract: Secret sharing and erasure coding based approaches have been used in distributed storage systems to ensure confidentiality, integrity, and availability of critical information. However, these approaches introduce some access overhead since each access now needs to access multiple data sites to retrieve the original data. In this paper, we investigate allocation of data partitions in the network for data objects that are partitioned by using secret sharing schemes or erasure coding schemes. We define the problem as finding M nodes in the network to host the data partitions so that the total communication cost (the sum of the read communication cost and write communication cost in terms of communication distance) is minimized to read K partitions and update M partitions by all nodes in the network. This problem is NP-hard. We propose a self-organizing feature map (SOFM)-like algorithm and a heuristic algorithm for this problem. The numerical study shows both algorithms are effective for this problem. [Copyright &y& Elsevier]

Abstract: In this paper, the delay-dependent state estimation problem for recurrent neural networks with both time-varying and distributed time-varying delays is investigated. Through available output measurements, a delay-dependent criterion is established to estimate the neuron states such that the dynamics of the estimation error is globally exponentially stable. The derivative of a time-varying delay satisfies and the activation functions are assumed to be neither monotonic nor differentiable, and more general than the recently commonly used Lipschitz conditions. Finally, two illustrative examples are given to demonstrate the usefulness of the obtained condition. [Copyright &y& Elsevier]

Abstract: The problem of robust stability for neural networks with time-varying delays and parameter uncertainties is investigated in this paper. The parameter uncertainties are described to be of linear fractional form, which include the norm bounded uncertainties as a special case. By introducing a new Lyapunov–Krasovskii functional and considering the additional useful terms when estimating the upper bound of the derivative of Lyapunov functional, new delay-dependent stability criteria are established in term of linear matrix inequality (LMI). It is shown that the obtained criteria can provide less conservative results than some existing ones. Numerical examples are given to demonstrate the applicability of the proposed approach. [Copyright &y& Elsevier]

Abstract: Many man-made objects are composed of a number of some simple geometric shapes such as lines, circles, rectangles, etc. Therefore, the detection of rectangular objects is an important issue to some practical applications such as the detection of buildings and vehicles in aerial imagery, the detection of license plates in car images, etc. Several methods have been proposed for solving the problem of the detection of rectangular objects. While some approaches are based on the detecting lines, some approaches are based on the Hough transform. Each approach has its own advantages and disadvantages (e.g., computational load). In this paper, we propose a class of neural networks with a special type of neural junctions for the detection of rectangular objects. The proposed neural networks can be trained in either an unsupervised mode or a batch mode. In contrast to some popular clustering algorithms such as the fuzzy -means algorithm and the fuzzy -rectangular shells algorithm, our approach is not based on minimizing an objective function but based on the idea of competitive learning. Based on the idea of competitive learning, the computational load can be decreased. Several data sets were tested to illustrate the effectiveness of our proposed approach. [Copyright &y& Elsevier]

Abstract: Automatic birdsong recognition using prediction-based singleton-type recurrent neural fuzzy networks (SRNFNs) is proposed in this paper. The recognition task consists of two stages. The first stage segments a significant portion from a birdsong sequence and the second stage performs recognition. For birdsong segmentation, an easy but effective segmentation approach based on time domain energy is proposed. For recognition, the linear predictive coding (LPC) coefficients of each frame in a segmented birdsong are extracted and used as features. These features are fed as inputs to SRNFN recognizers. The SRNFN is constructed by recurrent fuzzy if–then rules with fuzzy singletons in the consequences, and its recurrent aspect makes it suitable for processing patterns with temporal characteristics. In birdsong recognition, the sample prediction technique is used, where one SRNFN is responsible for learning the temporal birdsong relationships of only one species. The prediction error of each SRNFN is then used as a criterion for recognition. Experiments with 10 species of birds and their songs are performed, and a high recognition rate is achieved. Comparisons with a Takagi–Sugeno–Kang (TSK)-type recurrent fuzzy network (TRFN) and backpropagation neural network are also made in the experiments. [Copyright &y& Elsevier]

Abstract: Several implementations of Feedforward Neural Networks have been reported in scientific papers. These implementations do not allow the direct use of off-line trained networks. Usually, the problem is the lower precision (compared to the software used for training) or modifications in the activation function. In the present work, a hardware solution called Artificial Neural Network Processor, using a FPGA, fits the requirements for a direct implementation of Feedforward Neural Networks, because of the high precision and accurate activation function that were obtained. The resulting hardware solution is tested with data from a real system to confirm that it can correctly implement the models prepared off-line with MATLAB. [Copyright &y& Elsevier]

Abstract: This article presents a series of hardware implementations of a general regression neural network (GRNN) using FPGAs. The paper describes the study of this neural network using different fixed and floating point implementations. The implementation includes training as well as testing of the network. It is focused on precision loss and area and speed results of the resulting neural network coprocessor that can be used in a System on Programmable Chip. A GRNN is able to approximate functions and it has been used in control, prediction, fault diagnosis, engine management among others. They are mainly implemented as software entities because they require a great amount of complex mathematical operations. With the increasing power and capabilities of current FPGAs, now it is possible not only to translate them into hardware but, due to the reconfigurable feature of these devices, to explore different hardware/software partitions as well. These hardware implementations increase both the speed and performance of these neural networks and the designer can select the area–speed trade-off that best fits the application. [Copyright &y& Elsevier]

Abstract: In this paper, a model is considered to describe the dynamics of a class of non-autonomous neural networks with time-varying delays. By applying the properties of M-matrix, the techniques of inequality analysis and the Banach fixed point theorem, we obtain a series of new criteria on the dissipativity and existence of periodic attractor. Our results can extend and improve earlier ones. [Copyright &y& Elsevier]

A new load forecasting (LF) approach using bacterial foraging technique (BFT) trained wavelet neural network (WNN) is proposed in this paper. Artificial neural network (ANN) is combined with wavelet transform called wavelet neural network is applied for LF. The parameters of translation and dilation in the wavelet nodes and the weighting factors in the weighting nodes are tuned using BFT optimization. With the advantages of global search abilities of BFT as well as the multiresolution and localizing natures of wavelets, the networks are constructed which identifies the inherent non-linear characteristics of power system loads. The proposed approach is validated with Tamil Nadu Electricity Board (TNEB) system, India. The comparison of Delta Rule and BFT-based LF for different periods are depicted with their mean absolute percentage errors (MAPE). [Copyright &y& Elsevier]

Line flow or real-power contingency selection and ranking is performed to choose the contingencies that cause the worst overloading problems. In this paper, a cascade neural network-based approach is proposed for fast line flow contingency selection and ranking. The developed cascade neural network is a combination of a filter module and a ranking module. All the contingency cases are applied to the filter module, which is trained to classify them either in critical contingency class or in non-critical contingency class using a modified BP algorithm. The screened critical contingencies are passed to the ranking module (four-layered feed-forward artificial neural network (ANN)) for their further ranking. Effectiveness of the proposed ANN-based method is demonstrated by applying it for contingency screening and ranking at different loading conditions for IEEE 14-bus system. Once trained, the cascade neural network gives fast and accurate screening and ranking for unknown patterns and is found to be suitable for on-line applications at energy management centre. [Copyright &y& Elsevier]

Abstract: This paper proposes a new algorithm for a class of binary images thinning by using two PCNNs (pulse coupled neural networks). Once the travelling pulses of the two PCNNs meet, the thinning result is obtained. The criterion of pulses meeting is given, and the parameters of the PCNNs are also specified, which make the implementation of the proposed thinning algorithm easier. The algorithm is used to thin such a class of binary images, which separate the original images into two regions, as circularity-like images and ribbon-like shapes. Experimental results show that the proposed algorithm is efficient in extracting the skeleton of images (such as circularity-like images, ribbon-like shapes, handwriting ‘0’, etc.). [Copyright &y& Elsevier]

Abstract: Although the role of lateral inhibition in edge (contrast) enhancement is well known, it has not been parametrized. This paper investigates the imbalance between the lateral inhibitory and excitatory stimuli and its effect on the edge enhancement and stability. It is shown that this imbalance can be expressed through , a ratio of inhibitory to excitatory weights in a neuron. Stability requires to be less than the critical ratio . As approaches , edge enhancement increases, the rise being the sharpest just before instability. The increase in edge enhancement is also accompanied by an increase in the lateral spread of perturbations across the neuron layer. [Copyright &y& Elsevier]

This paper presents architectures for comparator trees capable of finding the minimum value of a large number of inputs. Such circuits are of general applicability although the intended application for which the circuits were designed is the winner-take-all function of a digital implementation of a neural network based on the self organising map. Mechanisms for reducing delay based on look-ahead logic within individual comparators and mechanisms based on multiplexor architectures of a comparator are compared for both propagation delay and area. [Copyright &y& Elsevier]

Problem decomposition and divide-and-conquer strategies have been proposed to improve the performance and realization of neural network solutions for complex problems. This paper reports on an experimental evaluation of performance gains brought about by problem decomposition for abductive network classifiers that classify four noisy waveform patterns having two waveform types (sine/cosine) and two different frequencies. Two-stage problem decomposition improves overall classification accuracy from 87.2% to 99%. Problem decomposition classifiers were found to be much more tolerant to model simplification and reduction in the training set size compared to monolithic solutions. This allows trading-off some of the large gain in classification performance for some other advantages that may be quite desirable in some applications, such as simpler models that execute faster and are easier to implement, smaller training sets, and shorter training times. A problem decomposition classifier is more accurate than a monolithic classifier in spite of the former being five times simpler, executing over two times faster, requiring one-fifth of the training data, and synthesized in one-eleventh of the training time. Performance is comparable with a neural network solution using the same decomposition method and significantly superior to an abductive network committee approach. [Copyright &y& Elsevier]

This paper suggests a decision support system for tactical air combat environment where not much prior information is available about the decision regions. We proposed a combination of unsupervised learning for clustering the data (to develop decision regions) and a feed forward neural network to classify the decision regions accurately. The clustered data is used as the inputs to the multi-layered feed forward neural network, which is trained using several higher order learning paradigms. Experiment results reveal that the proposed method is efficient. [Copyright &y& Elsevier]

Principal component analysis (PCA) is an effective method of linear dimensional reduction. Because of its simplicity in theory and implementation, it is often used for analyses in various disciplines. However, because of its linearity, PCA is not always suitable, and has redundancy in expressing data. To overcome this problem, some nonlinear PCA methods have been proposed. However, most of these methods have drawbacks, such that the number of principal components must be predetermined, and also the order of the generated principal components is not explicitly given. In this paper, we propose a nonlinear PCA algorithm that nonlinearly transforms data into principal components, and at the same time, preserving the order of the principal components, and we also propose a hierarchical neural network model to perform the algorithm. Moreover, our method does not need to know the number of principal components in advance. The effectiveness of the proposed model will be shown through experiments. [Copyright &y& Elsevier]

In this paper we show how genetic algorithms and neural networks are combined to build a high-performance signal pattern classifier (GNSPC). Signal patterns are intrinsic to many sensor-based systems. The goal of GNSPC is to differentiate among large numbers of signal pattern classes with low classification cost and high classification performance. Classification performance is measured by the correct classification of noisy signal patterns despite using pure signal patterns for building the classifier. GNSPC is basically a decision tree classifier with similarity classification rules. The rules are used to test the similarity of signal patterns. A combination of a genetic algorithm and a neural network is used to find the best rules for the decision tree. This combination provides powerful classification capabilities with great tuning flexibility for either performance or cost-efficiency. Learning techniques are employed to set the genetic algorithm global parameters and to obtain training data for the neural network. [Copyright &y& Elsevier]

In this paper, we propose a new clustering algorithm for a mixture of Gaussian-based neural network and self-growing probabilistic decision-based neural networks (SPDNN). The proposed self-growing cluster learning (SGCL) algorithm is able to find the natural number of prototypes based on a self-growing validity measure, Bayesian information criterion (BIC). The learning process starts from a single prototype randomly initialized in the feature space and grows adaptively during the learning process until most appropriate number of prototypes are found. We have conducted numerical and real-world experiments to demonstrate the effectiveness of the SGCL algorithm. In the results of using SGCL to train the SPDNN for data clustering and speaker identification problems, we have observed a noticeable improvement among various model-based or vector quantization-based classification schemes. [Copyright &y& Elsevier]

In this work we present the cooperative coevolution of multi-layer generalized perceptrons. This model is based on the cooperation of different subpopulations of modules, each one being a generalized multi-layered perceptron.In some previous works we have developed a modular cooperative coevolutive model for evolving multi-layer perceptrons with two hidden layers. This model performs very well but tends to generate big networks. In the present paper we show the results of substituting these multi-layer perceptrons by generalized multi-layer perceptrons, which allow a more compact representation of networks.The use of generalized multi-layer perceptrons improved the performance of the evolutionary model with regard to the evolution of other kinds of networks. Another improvement was that the networks obtained were much smaller. The comparison proved statistically significant by means of a Student''s t-test. [Copyright &y& Elsevier]