The identification of reliable biomarkers predicting outcomes in ultra-high risk (UHR) psychosis remains a vital challenge in preventive psychiatry. While genetic factors are implicated in psychosis risk, specific markers predicting social and cognitive trajectories have remained elusive. In this 24-month longitudinal study of UHR and Healthy Control (HC) participants, we investigated for the first time the predictive relationship between key immune-related genes, inflammatory response genes, stress response genes, and neurological-related genes with social and cognitive outcomes. Participants underwent comprehensive social and cognitive assessments at baseline and three 6-month intervals. We employed a brain-inspired Spiking Neural Network (SNN) model to map gene-behavior interactions and to identify key genetic predictors that influence both social and cognitive functioning. Additionally, we explore subgroup differences within the UHR population to further understand psychosis risk.

Conventional imaging devices often struggle to produce high-dynamic-range (HDR) images that accurately represent natural scenes. To overcome this limitation, multi-exposure image fusion (MEF) techniques have been introduced as a viable solution. Existing MEF approaches aim to enhance performance by optimizing or searching architectures. However, they face challenges in precise feature extraction and scene reconstruction, leading to distortion in the fused images. Additionally, most methods do not adequately address luminance variations across different image regions, which may result in the loss of essential details. To address these challenges, we present a novel luminance-aware MEF framework that integrates text-correlation priors (LarTap). By embedding textual information into fusion process, the proposed framework enhances content extraction and comprehension. Specifically, it consist of two key components: the text-image correlation network (N1) and the multi-exposure fusion network (N2). First, N1 performs correlation training to achieve a holistic alignment between text and image pairs. Its iterative vision encoders (VEs) generate text-correlated prior knowledge to facilitate the fusion process in N2. Second, N2 leverages these priors for scene reconstruction and dynamically adjusts luminance based on comparative perception. Extensive experiments on multiple datasets demonstrate that LarTap outperforms state-of-the-art methods. The source code is available at https://github.com/EnLong-wang/LarTap.

Multi-modality image fusion (MMIF) entails synthesizing images with detailed textures and prominent objects. Existing methods tend to use general feature extraction to handle different fusion tasks. However, these methods have difficulty breaking fusion barriers across various modalities owing to the lack of targeted learning routes. In this work, we propose a multi-scenario feature joint learning architecture, MLFuse, that employs the commonalities of multi-modality images to deconstruct the fusion progress. Specifically, we construct a cross-modal knowledge reinforcing network that adopts a multipath calibration strategy to promote information communication between different images. In addition, two professional networks are developed to maintain the salient and textural information of fusion results. The spatial-spectral domain optimizing network can learn the vital relationship of the source image context with the help of spatial attention and spectral attention. The edge-guided learning network utilizes the convolution operations of various receptive fields to capture image texture information. The desired fusion results are obtained by aggregating the outputs from the three networks. Extensive experiments demonstrate the superiority of MLFuse for infrared-visible image fusion and medical image fusion. The excellent results of downstream tasks (i.e., object detection and semantic segmentation) further verify the high-quality fusion performance of our method.

The Epileptor model is a mathematical framework utilized for simulating the transition from interictal to ictal local field potential (LFP) activity in the brain, with the aim of predicting and preventing epileptic seizures. This article introduces a novel approach integrating Lyapunov and Poincaré–Bendixson methods to analyze the stability of limit cycles in nonlinear systems, specifically focusing on Epileptors with a two-state dynamic. Our method accurately delineates the limit cycle boundary through eigenvalue-based analysis, facilitating precise assessment of stability properties and identification of critical regions linked to seizure initiation and termination. Through the investigation of the two-state dynamics of Epileptors, we gain deeper insights into the transition between low activity and seizure states, consequently improving our understanding of epileptic seizures. Our approach can be employed to establish stability conditions and determine the existence of limit cycles in Epileptor models, which can further aid in predicting and preventing epileptic seizures by identifying critical regions associated with seizure initiation and termination. The simulations conducted in this study demonstrate that the model under investigation exhibits stable limit cycle behavior and manifests bifurcation, with significant implications for the development of targeted interventions and more effective prediction and treatments for epilepsy. The findings indicate that the suggested approach establishes that external stimulation should not surpass 10.8 mA. Moreover, the initial normal state lies within the range of$-$1.6 to$-$0.1 ictal LFP. On the other hand, the LaSalle and eigenvalue methods individually cannot precisely determine the limit cycle region.

The serious concerns over the negative impacts of Deepfakes have attracted wide attentions in the community of multimedia forensics. The existing detection works achieve deepfake detection by improving the traditional backbone networks to capture subtle manipulation traces. However, there is no attempt to construct new backbone networks with different structures for Deepfake detection by improving the internal feature representation of convolution. In this work, we propose a novel Space-Frequency Interactive Convolution (SFIConv) to efficiently model the manipulation clues left by Deepfake. To obtain high-frequency features from tampering traces, a Multichannel Constrained Separable Convolution (MCSConv) is designed as the component of the proposed SFIConv, which learns space-frequency features via three stages, namely generation, interaction and fusion. In addition, SFIConv can replace the vanilla convolution in any backbone networks without changing the network structure. Extensive experimental results show that seamlessly equipping SFIConv into the backbone network greatly improves the accuracy for Deepfake detection. In addition, the space-frequency interaction mechanism does benefit to capturing common artifact features, thus achieving better results in cross-dataset evaluation. Our code will be available athttps://github.com/EricGzq/SFIConv.

Sand-dust videos obtained in a low-light environment are characterized by low contrast, nonuniform illumination, color cast, and considerable noise. To realize sand-dust removal and brightness enhancement simultaneously, this article proposes an online low-light sand-dust video enhancement method using adaptive dynamic brightness correction and a rolling guidance filter. The proposed dual-threshold interframe detection strategy involves two methods to treat low-light sand-dust video frames. The first method involves two components: an adaptive dynamic brightness correction algorithm to correct the color deviation of the low-light video frame and improve its brightness and a rolling guidance filter combined with guided image filtering to enhance the frame details. The second method enhances the quality of the incoming frame by reducing the amount of calculation. The first frame of the video is processed using the first method. The processing method of each subsequent frame is determined according to its interframe detection value with the buffer frame. Through qualitative and quantitative comprehensive experiments on low-light sand-dust images and videos, the performance of the proposed method is compared with those of state-of-the-art methods. The proposed method for frame quality improvement achieves the best visual effect in enhancing the quality of low-light sand-dust images, as indicated by the best objective evaluation indicators. Moreover, compared with the framewise enhancement method, the video processing efficiency associated with the dual-threshold interframe detection strategy is 2.77 times higher.

At present, multimodal medical image fusion technology has become an essential means for researchers and doctors to predict diseases and study pathology. Nevertheless, how to reserve more unique features from different modal source images on the premise of ensuring time efficiency is a tricky problem. To handle this issue, we propose a flexible semantic-guided architecture with a mask-optimized framework in an end-to-end manner, termed as GeSeNet. Specifically, a region mask module is devised to deepen the learning of important information while pruning redundant computation for reducing the runtime. An edge enhancement module and a global refinement module are presented to modify the extracted features for boosting the edge textures and adjusting overall visual performance. In addition, we introduce a semantic module that is cascaded with the proposed fusion network to deliver semantic information into our generated results. Sufficient qualitative and quantitative comparative experiments (i.e., MRI-CT, MRI-PET, and MRI-SPECT) are deployed between our proposed method and ten state-of-the-art methods, which shows our generated images lead the way. Moreover, we also conduct operational efficiency comparisons and ablation experiments to prove that our proposed method can perform excellently in the field of multimodal medical image fusion. The code is available at https://github.com/lok-18/GeSeNet.

In this paper, we present a novel Liquid State Machine (LSM) based approach for modelling of multimodal longitudinal data: the Mosaic LSM. Our model harnesses the strengths of multiple LSMs, each designed to capture the temporal patterns of a specific data modality. This temporal information is then added to the raw data to create a composite representation that encompasses both the multimodal and the longitudinal aspects of the data. We demonstrate the performance of our approach on a real-world dataset that contains clinical, cognitive, and genetic modalities with the aim of predicting the Ultra-High Risk (UHR) status in individuals, six months in advance. Our results show that the Mosaic LSM outperforms traditional machine learning models, achieving an outstanding Matthew's Correlation Coefficient of 0.84 and prediction accuracy of 92.4%. Overall, our work highlights the potential of Mosaic LSM as a powerful tool for disease prognosis, and its ability to leverage both the multimodality and temporality of the data to improve performance.

The paper proposes a novel hierarchical recurrent neural network architecture for on-line classification and interpretation of EEG data. It incorporates two dynamic pools of neurons - one based on NeuCube three dimensional structure of spiking neurons, spatially mapping a brain template and connected via spike-timing dependent plastic synapses and another Echo state neural network (ESN) reservoir of sparsely connected hyperbolic tangent neurons that is able to learn on-line to classify continuously extracted from the Cube spike-rate features. The aim of the work was to interpret and classify in a brain-inspired manner dynamic spatio-temporal brain signals. The achieved results demonstrate improved classification accuracy on a benchmark EEG data set along with a good interpretability of the data. In future, the proposed method can be used for classification of other brain spatio-temporal data, such as ECOG and fMRI.

Selecting informative features, such as accurate biomarkers for disease diagnosis, prognosis and response to treatment, is an essential task in the field of bioinformatics. Medical data often contain thousands of features and identifying potential biomarkers is challenging due to small number of samples in the data, method dependence and non-reproducibility. This paper proposes a novel ensemble feature selection method, named Filter and Wrapper Stacking Ensemble (FWSE), to identify reproducible biomarkers from high-dimensional omics data. In FWSE, filter feature selection methods are run on numerous subsets of the data to eliminate irrelevant features, and then wrapper feature selection methods are applied to rank the top features. The method was validated on four high-dimensional medical datasets related to mental illnesses and cancer. The results indicate that the features selected by FWSE are stable and statistically more significant than the ones obtained by existing methods while also demonstrating biological relevance. Furthermore, FWSE is a generic method, applicable to various high-dimensional datasets in the fields of machine intelligence and bioinformatics.

Nowadays, deep learning has made rapid progress in the field of multi-exposure image fusion. However, it is still challenging to extract available features while retaining texture details and color. To address this difficult issue, in this paper, we propose a coordinated learning network for detail-refinement in an end-to-end manner. Firstly, we obtain shallow feature maps from extreme over/under-exposed source images by a collaborative extraction module. Secondly, smooth attention weight maps are generated under the guidance of a self-attention module, which can draw a global connection to correlate patches in different locations. With the cooperation of the two aforementioned used modules, our proposed network can obtain a coarse fused image. Moreover, by assisting with an edge revision module, edge details of fused results are refined and noise is suppressed effectively. We conduct subjective qualitative and objective quantitative comparisons between the proposed method and twelve state-of-the-art methods on two available public datasets, respectively. The results show that our fused images significantly outperform others in visual effects and evaluation metrics. In addition, we also perform ablation experiments to verify the function and effectiveness of each module in our proposed method. The source code can be achieved athttps://github.com/lok-18/LCNDR.

In low-illumination environments such as at night, due to factors such as the large monitoring field of an eagle eye, short sensor exposure time, and high-density random noise, the video images collected by image sensors generally have poor visual quality and low signal-to-noise ratio, which makes it difficult for surveillance systems to detect weak changes. To solve this problem, we propose a method for image change detection (CD) in surveillance video based on optimized k-medoids clustering and adaptive fusion of difference images (DIs). First, for the input multitemporal video surveillance images, two DIs are obtained by log-ratio and extremum pixel ratio operators. Then, the two DIs are adaptively fused by combining the local energy of DIs and the Laplacian pyramid. Simultaneously, the fused DI is compressed by the normalization function, and the final DI is obtained via the improved adaptive median filter. Finally, the changed image is obtained by using the optimized k-medoids clustering algorithm. The experimental results show that the proposed method can accurately and effectively detect weak changes in the eagle eye surveillance picture in a low-illumination environment. Compared with those of other methods, the accuracy and robustness of the proposed method are higher, and the running time of the algorithm is shorter. Moreover, it will not generate a false alarm due to the influence of noise in unchanged scenes.

The article demonstrates for the first time that a brain-inspired spiking neural network (SNN) architecture can be used not only to learn spatio-temporal data, but also to extract fuzzy spatio-temporal rules from such data and to update these rules incrementally in a transfer learning mode. We propose a method, where a SNN model learns incrementally new time-space data related to new classes/tasks/categories, always utilizing some previously learned knowledge, and presents the evolved knowledge as fuzzy spatio-temporal rules. Similarly, to how the brain manifests transfer learning, these SNN models do not need to be restricted in number of layers and neurons in each layer as they adopt self-organizing learning principles. The continuously evolved fuzzy rules from spatio-temporal data are interpretable for a better understanding of the processes that generate the data. The proposed method is based on a brain-inspired SNN architecture NeuCube, which is structured according to a brain three-dimensional structural template. It is illustrated on tasks of incremental and transfer learning and knowledge transfer using spatio-temporal data measuring brain activity, when subjects are performing tasks in space and time. The method is a general one and opens the field to create new types of adaptable and explainable spatio-temporal learning systems across domain areas.

Fuzzy clustering is an essential algorithm in image segmentation, and most of them are based on fuzzy c-mean algorithms. However, it is sensitive to noise, center point selection, cluster number, and distance metric. To address this problem, we propose a new fuzzy clustering method based on low-rank representation (LRR) for image segmentation, which integrates low-rank structure with fuzzy theory. First, we improve the morphological reconstruction superpixel method based on edge detection by introducing anisotropy to enhance the image edge. Thus, on the one hand, the improved morphological reconstruction superpixel method can improve its noise-resistance performance; on the other hand, the complexity of the subsequent low-rank computation can be reduced by enhancing the superpixels constructed by the edges. Second, inspired by the fact that rank can represent correlation, we propose the concept of fuzzy low-rank structure, which is not dealing with data directly but with the relationship between data. Specifically, we perform rank minimization on the constructed membership matrix to obtain the optimal matrix. To obtain better clustering results, we added the Frobenius norm of the fuzzy matrix as a fuzzy regularization term in the LRR model to achieve global convergence and obtain a membership matrix with a strong element correlation. Finally, we obtain the final clustering results by clustering the processed membership matrix using a subspace clustering with a low-rank structure constraint. Experiments performed on artificial and real-world images show that the proposed method is more effective and efficient than the current state-of-the-art methods.

Gaze estimation plays an essential role in human attention recognition, human behavior analysis and augmented reality applications. Most of the deep neural network-based gaze estimation techniques apply supervised learning to extract features and regress 3D gaze vectors directly, leading to a vulnerability of high labor cost and limited generalization. In this work, we proposed a weakly-supervised method to jointly optimize the depth values of eye landmarks and relative poses with a multi-view geometric constraint to determine the final gaze vectors of observers. Specifically, we feed in sequential eye region images, and design a depth regression network to estimate the depth of the eye region landmarks, which are further utilized by the pose estimation network to estimate the relative changes of gaze vectors with multi-view geometric constraints in the iris regions. Experiments on both synthetic and real data show that the proposed method is feasible and promising to learn gaze estimation without strong pose supervision.

What algorithms to choose for an incremental, predictive, and more importantly, explainable learning and modelling of time series data, and specifically – economic and financial time series? Will these algorithms reveal and explain abnormality in time series over time? This problem is part of the challenges in the area of explainable AI and life-long learning systems. Three widely used evolving connectionist systems (ECOS) that offer a solution to the above problems, are compared in the paper. The first two models, EFuNN and DENFIS, are neuro-fuzzy models that deal with vector-based data, while the spiking neural network model NeuCube deals with temporal and spatio-temporal data. The case study data used to demonstrate the qualities of these techniques is time series data related to Bulgarian petroleum oil imports from various trading partners. Graphical visualization enhances the deep analysis, pattern recognition, and knowledge extraction from the models. Conclusions are drawn in the sense that each of these techniques reveals different aspects of the data and the problem in hand and a single model solution of a complex problem is never going to be complete. Future work outlines integration of time series and on-line news to achieve a better predictive accuracy and a better understanding of the data.

Compared with single-band images, multispectral images with multiple wavelengths contain more spectral information and can fully reflect the detailed features of ground objects in different bands. To synthesize the feature information, a method of multispectral image enhancement based on improved weighted principal component analysis (WPCA) and improved fractional differential (IFD) filtering is proposed. First, the optimum index factor (OIF) model is modified to select the bands for easy follow-up processing. Then an improved WPCA transform that uses the average gradient and texture roughness is applied to compensate for each band. This approach preserves the main information, compresses the amount of image data, and obtains uncorrelated principal components. According to the correlation of neighboring pixels, a new mask is introduced to enhance the first principal component. Finally, the brightness values of the image are adjusted after inverse WPCA transform to obtain the final enhanced image. The experimental results demonstrate the superiority of the proposed method over related methods. Its future practical applications are discussed in the conclusion.

This paper is in the scope of emotion recognition by employing a recurrent spiking neural network (BI-SNN) architecture for modelling, mapping, learning, classifying, visualising, and understanding of spatio-temporal Electroencephalogram (EEG) data related to different emotional states. It further explores, develops, and applies a methodology based on the NeuCube BI-SNN, that includes methods for EEG data encoding, data mapping into a 3-dimensional BI-SNN model, unsupervised learning using spike-timing dependent plasticity (STDP) rule, spike-driven supervised learning, output classification, network analysis, and model visualisation and interpretation. The research conducted to model different emotional subtypes through mapping both space (brain regions) and time (brain dynamics) components of EEG brain data into SNN architecture. Here, a benchmark EEG dataset was used to design an empirical study that consisted of different experiments for classification of emotions. The obtained accuracy of 94.83% for EEG classification of four types of emotions was superior when compared with traditional machine learning techniques. The BI-SNN models not only detected the brain activity patterns related to positive and negative emotions with a high accuracy, but also revealed new knowledge about the brain areas activated in relation to different emotions. The research confirmed that neural activation increased in the frontal sites of brain (F7, F3, AF4) associated with positive emotions, while in the case of the negative emotions, connectivity strength was concentrated in the frontal (F4, AF3, F7, F8) and parietal sites of the brain (P7, P8).

Micro-expression can reveal underlying genuine emotions, but those rapid and subtle changes are hard to be captured by humans. Most existing research focuses on frontal face micro-expression recognition, which largely prevents the developed methods from the real applications and ignores the underlying geometry information. In this paper, we propose a multiview geometry consistency framework to enable the same emotion to be recognized under different perspectives, which is difficult for existing systems. Based on the developed 3D face reconstruction network, the multi-view micro-expression recognition framework empowers the emotion recognition capability to learn from multiple perspectives of the 3D reconstructed faces based on view-consistency, and a spiking neural network is further applied to capture omitted tiny and detailed changes. With a sequence of images, we explore the subtle changes across frames through optical flow, which, as a clue, enhances the performance of our designated network for micro-expression recognition. Extensive experiments on benchmark micro-expression datasets CAS(ME)2and SMIC demonstrate the proposed method achieves promising results on novel-view micro-expression recognition where existing methods mainly fail.

This paper studies sleep stage classification using NeuCube, a Spiking Neural Network (SNN) architecture, simulated on SpiNNaker, a neuromorphic computer. The sleep electroencephalogram (EEG) time series is converted to spikes and provided as an input to NeuCube. Relevant feature vectors are extracted at different stages of training. We used six standard machine learning classifiers on different combinations of these feature vectors and calculated 5-fold cross-validation accuracy. We observed that the gradient boosted decision trees classifier performed the best by achieving 81.25% accuracy on a combination of two feature vectors. An evaluation of the results using confusion matrices and classification reports showed that the Awake, N2, SWS and REM sleep stages can be classified with ≥ 80% F1-score using the gradient boosted decision trees algorithm. Overall, our proof-of-concept work towards autonomous sleep-stage classification using NeuCube shows promise and will form the base for continued research in this direction.

Applications that generate huge amounts of data in the form of fast streams are becoming increasingly prevalent, being therefore necessary to learn in an online manner. These conditions usually impose memory and processing time restrictions, and they often turn into evolving environments where a change may affect the input data distribution. Such a change causes that predictive models trained over these stream data become obsolete and do not adapt suitably to new distributions. Specially in these non-stationary scenarios, there is a pressing need for new algorithms that adapt to these changes as fast as possible, while maintaining good performance scores. Unfortunately, most off-the-shelf classification models need to be retrained if they are used in changing environments, and fail to scale properly. Spiking Neural Networks have revealed themselves as one of the most successful approaches to model the behavior and learning potential of the brain, and exploit them to undertake practical online learning tasks. Besides, some specific flavors of Spiking Neural Networks can overcome the necessity of retraining after a drift occurs. This work intends to merge both fields by serving as a comprehensive overview, motivating further developments that embrace Spiking Neural Networks for online learning scenarios, and being a friendly entry point for non-experts.

This article proposes a new spike encoding and decoding algorithm for analog data. The algorithm uses the pulsewidth modulation principles to achieve a high reconstruction accuracy of the signal, along with a high level of data compression. Two benchmark data sets are used to illustrate the method: stock index time series and human voice data. Applications of the method for spiking neural network (SNN) modeling and neuromorphic implementations are discussed. The proposed method would allow the development of new applications of SNNs as regression techniques for predictive time-series modeling.

Spiking neural networks (SNNs) receive trains of spiking events as inputs. In order to design efficient SNN systems, real-valued signals must be optimally encoded into spike trains so that the task-relevant information is retained. This paper provides a systematic quantitative and qualitative analysis and guidelines for optimal temporal encoding. It proposes a methodology of a three-step encoding workflow: method selection by signal characteristics, parameter optimization by error metrics between original and reconstructed signals, and validation by comparison of the original signal and the encoded spike train. Four encoding methods are analyzed: one stimulus estimation [Ben's Spiker algorithm (BSA)] and three temporal contrast [threshold-based, step-forward (SW), and moving-window (MW)] encodings. A short theoretical analysis is provided, and the extended quantitative analysis is carried out applying four types of test signals: step-wise signal, smooth (sinusoid) signal with added noise, trended smooth signal, and event-like smooth signal. Various time-domain and frequency spectrum properties are explored, and a comparison is provided. BSA, the only method providing unipolar spikes, was shown to be ineffective for step-wise signals, but it can follow smoothly changing signals if filter coefficients are scaled appropriately. Producing bipolar (positive and negative) spike trains, SW encoding was most effective for all types of signals as it proved to be robust and easy to optimize. Signal-to-noise ratio (SNR) can be recommended as the error metric for parameter optimization. Currently, only a visual check is available for final validation.

Objective:Due to the non-stationarity and high trialto-trial variability, online event prediction from biomedical signals is challenging. This is more significant when it is applied to neurological rehabilitation where the person incrementally learns to regain the control of movement. eSPANNet is a computational model inspired by the incremental learning for motor control in living nervous systems. It is inspired by the concept of 'population vectors' which have been experimentally proven by several computational neuroscience studies. In this paper, we present a proof-of-concept study on the proposed computational model. Our goal is to utilize the polychronization effect of Spiking Neural Networks to develop a better neural decoder for Brain-Computer Interfaces. Methods: The eSPANNet model contains a network of Spike Pattern Association Neurons, a spiking neuron model which is able to emit spikes at the desired time-point. Results: The proposed approach was experimentally validated using the finger flexion prediction dataset from the fourth BCI competition. The results show that eSPANNet results in 1) a higher classification accuracy, sensitivity and F1 score compared to several other multi-class classifiers and, 2) a better approximation of the actual movement compared to several regression analysis based approaches. Conclusion and Significance: The novelty of our algorithm is the ability to learn which inputs to focus on in an online manner. We suggest that the eSPANNet is a better BCI decoder due to its i) incremental and life-long learning, ii) compatibility with the neuromorphic platforms and, iii) ability to address the non-stationarity of brain data.

Limb amputation is a global problem. Prosthetic limbs can enhance the quality of life of amputees. To this end, anthropomorphic design and intuitive manipulation are two essential requirements. This paper presents a motor control framework for prosthetic control through Brain-Machine Interface (BMI) using Finite Automata Theory, and NeuCube Evolving Spiking Neural Network (SNN) architecture. Voluntary control of prosthetics requires decoding motor commands from the Central Nervous System of the amputee. Selection of the most suitable biomedical signal depends on many parameters such as level of amputation and muscle atrophy. Non-invasive BMI's have the possibility of supporting a wider range of amputees as it extracts the motor commands from the brain. In this paper, we present a proof of concept study on whether a cognitive computational model that is inspired by the motor control of the human body through muscle synergies combined with an anthropomorphic mechanical design, can result in accurate and robust prosthetic control through a noninvasive BMI. In future, learning of a complex Finite Automata that reflects complex upper limb motor behaviours will be investigated.

This paper proposes a method utilising spiking neural networks (SNN) for modelling, visualising and comparing the brain data under complex mental states. The method was applied to a cognitive task performed by 23 participants while they were making decision on a moral dilemma situation-related task. An SNN evolving spatiotemporal data architecture is used to learn and visualise the neural activity across different brain regions. The model developed allows for studying the patterns of electrical activity of neurons elicited during complex decision making processes such as moral-related tasks. This could be used for predictive analysis of various aspects of human behavior during decision making and for other related cognitive tasks.

The purpose of this paper was to investigate a pipeline for processing temporal gene expression data using spiking neural networks and temporal feature selection techniques that would allow for genomic marker discovery. A promising temporal feature selection method was tested using the NeuCube for classification against a set of previously identified genes using a dataset from Ebola vaccine trials. Classification results from the temporal selection method and the NeuCube model were significantly better than when using previously published gene sets. The discovered gene markers and their corresponding gene interaction network (GIN) are also new and have not been published before. This demonstrates both the potential of the examined feature selection method, and how Spiking Neural Networks (SNN) can be used for time series modelling and the discovery of novel GIN's. Future work includes improving temporal feature selection methods for gene expression data, and refining the use of SNN's for time series analysis.

Discriminative Correlation Filters (DCFs)-based approaches have recently achieved competitive performance in visual tracking. However, such conventional DCF-based trackers often lack the discriminative ability due to the shallow architecture. As a result, they can hardly tackle drastic appearance variations and easily drift when the target suffers heavy occlusions. To address this issue, a novel densely connected DCFs framework is proposed for visual tracking. We incorporate multiple nested DCFs into the deep learning architecture, and then train the compact network with the data-specific target. Specifically, feature maps and interim response maps are shared and reused throughout the whole network. By doing so, the implicit information carried out by each DCF is fully exploited to enhance the model representation ability during the tracking process. Moreover, a multiscale estimation scheme is developed to account for scale variations. Experimental results on the benchmarks demonstrate that the proposed approach achieves outstanding performance compared to the existing state-of-the-art trackers.

Correlation filter-based trackers (CFTs) have recently raised considerable attention in visual tracking and achieved competitive performance. Nevertheless, conventional structures of such CFTs build a shallow architecture with a single correlation filter, which cannot comprehensively depict the target appearance. Hence, these trackers lack strong discriminative ability and easily drift when the target suffers drastic appearance variations. To address the limitations, we propose Dirac-weighted cascading correlation filters (DWCCF) for visual tracking. It incorporates cascading characteristics of multiple filters to construct the target appearance model, and dynamically learns Dirac weights for each filter, which is accordingly robust to appearance variations. Besides, we design a boundary penalization strategy to adaptively reduce the boundary effects, which efficiently improves the detection precision for tracking. Qualitative and quantitative evaluations on OTB-2013 and OTB-2015 datasets demonstrate that the proposed DWCCF significantly outperforms other state-of-the-art methods.

Recent progress in a noninvasive brain data sampling technology has facilitated simultaneous sampling of multiple modalities of brain data, such as functional magnetic resonance imaging, electroencephalography, diffusion tensor imaging, and so on. In spite of the potential benefits from integrating predictive modeling of multiple modality brain data, this area of research remains mostly unexplored due to a lack of methodological advancements. The difficulty in fusing multiple modalities of brain data within a single model lies in the heterogeneous temporal and spatial characteristics of the data sources. Recent advances in spiking neural network systems, however, provide the flexibility to incorporate multidimensional information within the model. This paper proposes a novel, unsupervised learning algorithm for fusing temporal, spatial, and orientation information in a spiking neural network architecture that could potentially be used to understand and perform predictive modeling using multimodal data. The proposed algorithm is evaluated both qualitatively and quantitatively using synthetically generated data to characterize its behavior and its ability to utilize spatial, temporal, and orientation information within the model. This leads to improved pattern recognition capabilities and performance along with robust interpretability of the brain data. Furthermore, a case study is presented, which aims to build a computational model that discriminates between people with schizophrenia who respond or do not respond to monotherapy with the antipsychotic clozapine.

This paper introduces a new methodology for dynamic learning, visualization, and classification of functional magnetic resonance imaging (fMRI) as spatiotemporal brain data. The method is based on an evolving spatiotemporal data machine of evolving spiking neural networks (SNNs) exemplified by the NeuCube architecture [1]. The method consists of several steps: mapping spatial coordinates of fMRI data into a 3-D SNN cube (SNNc) that represents a brain template; input data transformation into trains of spikes; deep, unsupervised learning in the 3-D SNNc of spatiotemporal patterns from data; supervised learning in an evolving SNN classifier; parameter optimization; and 3-D visualization and model interpretation. Two benchmark case study problems and data are used to illustrate the proposed methodology-fMRI data collected from subjects when reading affirmative or negative sentences and another one-on reading a sentence or seeing a picture. The learned connections in the SNNc represent dynamic spatiotemporal relationships derived from the fMRI data. They can reveal new information about the brain functions under different conditions. The proposed methodology allows for the first time to analyze dynamic functional and structural connectivity of a learned SNN model from fMRI data. This can be used for a better understanding of brain activities and also for online generation of appropriate neurofeedback to subjects for improved brain functions. For example, in this paper, tracing the 3-D SNN model connectivity enabled us for the first time to capture prominent brain functional pathways evoked in language comprehension. We found stronger spatiotemporal interaction between left dorsolateral prefrontal cortex and left temporal while reading a negated sentence. This observation is obviously distinguishable from the patterns generated by either reading affirmative sentences or seeing pictures. The proposed NeuCube-based methodology offers also a superior classification accuracy when compared with traditional AI and statistical methods. The created NeuCube-based models of fMRI data are directly and efficiently implementable on high performance and low energy consumption neuromorphic platforms for real-time applications.

This paper proposes a new method for an optimized mapping of temporal variables, describing a temporal stream data, into the recently proposed NeuCube spiking neural network (SNN) architecture. This optimized mapping extends the use of the NeuCube, which was initially designed for spatiotemporal brain data, to work on arbitrary stream data and to achieve a better accuracy of temporal pattern recognition, a better and earlier event prediction, and a better understanding of complex temporal stream data through visualization of the NeuCube connectivity. The effect of the new mapping is demonstrated on three benchmark problems. The first one is the early prediction of patient sleep stage event from temporal physiological data. The second one is the pattern recognition of dynamic temporal patterns of traffic in the Bay Area of California and the last one is the Challenge 2012 contest data set. In all the cases, the use of the proposed mapping leads to an improved accuracy of pattern recognition and event prediction and a better understanding of the data when compared with traditional machine learning techniques or SNN reservoirs with an arbitrary mapping of the variables.

This study analyses different representations of large spiking neural network (SNN) structures for conventional computers and uses the NeuCube SNN architecture as a case study. The representation includes neuronal connectivity and network's and neurons' states during the learning process. Three different structure types, namely adjacency matrix, adjacency list, and edge-weight table, were compared in terms of their storage needs and execution time performance of a learning algorithm, for varying numbers of neurons in the network. Comparative analysis shows that the adjacency list, combined with a backwards indexing mechanism, scales up most efficiently both in terms of performance and of storage requirements. The optimal algorithm was further used to simulate a large scale NeuCube system with 241,606 spiking neurons in a 3D space for prediction and analysis of benchmark spatio-temporal data.

Motivated by the dramatic rise of neurological disorders, we propose a SNN technique to model electroen-cephalography (EEG) data collected from people affected by Alzheimer's Disease (AD) and people diagnosed with mild cognitive impairment (MCI). An evolving spatio-temporal data machine (eSTDM), named the NeuCube architecture, is used to analyse changes of neural activity across different brain regions. The model developed allows for studying AD progression and for predicting whether a patient diagnosed with MCI is more likely to develop AD.

This paper introduces a novel personalised modelling framework and system for analysing Spatio-Temporal Brain Data (STBD) along with person clinical static data. For every individual, based on selected subset of similar to this individual clinical data, a subset of STBD is used for training a personalised Spiking Neural Network (PSNN) model using the recently proposed NeuCube SNN architecture. The proposed method is illustrated on a case study of personalised modelling using clinical and EEG data of two groups of subjects - drug addicts and addicts under medication. The PSNN models help to achieve a better classification accuracy compared to global SNN models or when using traditional AI methods. A PSNN model visualisation enables discovery of new knowledge about individual persons and to distinguish complex STBD across subjects.

Large-scale cancer genomics projects are providing a wealth of somatic mutation data from a large number of cancer patients. However, it is difficult to obtain several samples with a temporal order from one patient in evaluating the cancer progression. Therefore, one of the most challenging problems arising from the data is to infer the temporal order of mutations across many patients. To solve the problem efficiently, we present a Network-based method (NetInf) to Infer cancer progression at the pathway level from cross-sectional data across many patients, leveraging on the exclusive property of driver mutations within a pathway and the property of linear progression between pathways. To assess the robustness of NetInf, we apply it on simulated data with the addition of different levels of noise. To verify the performance of NetInf, we apply it to analyze somatic mutation data from three real cancer studies with large number of samples. Experimental results reveal that the pathways detected by NetInf show significant enrichment. Our method reduces computational complexity by constructing gene networks without assigning the number of pathways, which also provides new insights on the temporal order of somatic mutations at the pathway level rather than at the gene level.

This paper presents spiking neural networks (SNNs) for remote sensing spatiotemporal analysis of image time series, which make use of the highly parallel and low-power-consuming neuromorphic hardware platforms possible. This paper illustrates this concept with the introduction of the first SNN computational model for crop yield estimation from normalized difference vegetation index image time series. It presents the development and testing of a methodological framework which utilizes the spatial accumulation of time series of Moderate Resolution Imaging Spectroradiometer 250-m resolution data and historical crop yield data to train an SNN to make timely prediction of crop yield. The research work also includes an analysis on the optimum number of features needed to optimize the results from our experimental data set. The proposed approach was applied to estimate the winter wheat (Triticum aestivum L.) yield in Shandong province, one of the main winter-wheat-growing regions of China. Our method was able to predict the yield around six weeks before harvest with a very high accuracy. Our methodology provided an average accuracy of 95.64%, with an average error of prediction of 0.236 t/ha and correlation coefficient of 0.801 based on a nine-feature model.

Epilepsy is the most diffuse brain disorder that can affect people's lives even on its early stage. In this paper, we used for the first time the spiking neural networks (SNN) framework called NeuCube for the analysis of electroencephalography (EEG) data recorded from a person affected by Absence Epileptic (AE), using permutation entropy (PE) features. Our results demonstrated that the methodology constitutes a valuable tool for the analysis and understanding of functional changes in the brain in term of its spiking activity and connectivity. Future applications of the model aim at personalised modelling of epileptic data for the analysis and the event prediction.

sEMG (surface electromyography) signals have been used as human-machine interface to control robots or prostheses in recent years. sEMG-based torque estimation is a widely research methodology to obtain human motion intention. Most researches focus on improving the accuracy of sEMG-torque models, which often makes them complicated and confined in the laboratory research. However, an accurate estimation of muscle torque could be unnecessary to perform the robot-assisted rehabilitation training. This paper proposes a practical method to estimate the net muscle torques of lower limbs using sEMG, which can be used to implement a real-time coordinated active training with iLeg-a horizontal exoskeleton for lower limb rehabilitation developed at our laboratory. Two three-layer back propagation (BP) neural networks are built to estimate the net muscle torques at hip and knee joints respectively. Experimental results show that the well-trained neural networks estimate the user's motion intention in real-time, and can assist the user to perform an active training with iLeg.

BACKGROUND: Large-scale cancer genomic projects are providing lots of data on genomic, epigenomic and gene expression aberrations in many cancer types. One key challenge is to detect functional driver pathways and to filter out nonfunctional passenger genes in cancer genomics. Vandin et al. introduced the Maximum Weight Sub-matrix Problem to find driver pathways and showed that it is an NP-hard problem. METHODS: To find a better solution and solve the problem more efficiently, we present a network-based method (NBM) to detect overlapping driver pathways automatically. This algorithm can directly find driver pathways or gene sets de novo from somatic mutation data utilizing two combinatorial properties, high coverage and high exclusivity, without any prior information. We firstly construct gene networks based on the approximate exclusivity between each pair of genes using somatic mutation data from many cancer patients. Secondly, we present a new greedy strategy to add or remove genes for obtaining overlapping gene sets with driver mutations according to the properties of high exclusivity and high coverage. RESULTS: To assess the efficiency of the proposed NBM, we apply the method on simulated data and compare results obtained from the NBM, RME, Dendrix and Multi-Dendrix. NBM obtains optimal results in less than nine seconds on a conventional computer and the time complexity is much less than the three other methods. To further verify the performance of NBM, we apply the method to analyze somatic mutation data from five real biological data sets such as the mutation profiles of 90 glioblastoma tumor samples and 163 lung carcinoma samples. NBM detects groups of genes which overlap with known pathways, including P53, RB and RTK/RAS/PI(3)K signaling pathways. New gene sets with p-value less than 1e-3 are found from the somatic mutation data. CONCLUSIONS: NBM can detect more biologically relevant gene sets. Results show that NBM outperforms other algorithms for detecting driver pathways or gene sets. Further research will be conducted with the use of novel machine learning techniques.

This research presents a generic framework and methods for mining temporal rules from multiple time-series data and its application to ecological data. The aphids dataset that tracks the trajectory of aphid infestations over time has been well researched in a number of studies. Those studies concentrated on predicting the scale of infestation over time. The focus of our research is to identify environmental factors that predict, in a temporal fashion, high incidence of aphid activity. This required the development of a novel framework for knowledge extraction from multiple time-series data and a method for discretization of numeric data as well-known methods such as SAX did not perform adequately due to the non-Gaussian nature of the data involved. Our experimentation yielded new insights into the environmental factors that may influence pest outbreak which are captured in the form of simple actionable rules that would be of interest to the farming community.

This paper is a continuation of previous published work by the same authors on Personalized Modelling and Evolving Spiking Neural Network Reservoir architecture (PMeSNNr). The focus is on improvement of predictive modeling methods for the stroke occurrences case study utilizing an enhanced NeuCube architecture. The adaptability of the new architecture leads towards understanding feature correlations that affect the outcome of the study and extracts new knowledge from hidden patterns that reside within the associations. Through this new method, estimation of the earliest time point for stroke prediction is possible. This study also highlighted the improvement from designing a new experimental dataset compared to previous experiments. Comparative experiments were also carried out using conventional machine learning algorithms such as kNN, wkNN, SVM and MLP to prove that our approach can result in much better accuracy level.

The paper is a feasibility analysis of using the recently introduced by one of the authors spiking neural networks architecture NeuCube for modelling and recognition of complex EEG spatio-temporal data related to both physical and intentional (imagined) movements. The preliminary experiments reported in the paper suggest that NeuCube is much more efficient for the task than standard machine learning techniques, resulting in high recognition accuracy, a better adaptability to new data, a better interpretation of the models, leading to a better understanding of the brain data and the processes that generated it.

Early event prediction challenges most of existing modeling methods especially when dealing with complex spatio-temporal data. In this paper we propose a new method for predictive data modelling based on a new development of the recently proposed NeuCube spiking neural network architecture, called here NeuCube(ST). The NeuCube uses a Spiking Neural Network reservoir (SNNr) and dynamic evolving Spiking Neuron Network (deSNN) classifier. NeuCube(ST)is an integrated environment including data conversion into spike trains, input variable mapping, unsupervised learning in the SNNr, supervised classification learning, activity visualization and network structure analysis. A case study on a real world ecological data set is presented to demonstrate the validity of the proposed method.

SUMMARY The level set method has been widely used in image segmentation; however, the complexity of the computation has restricted its application field. Also, it is a big challenge to segment remote sensing image mainly because of the complex terrain. In this paper, an enhanced multiphase phase level set method based on the Chan–Vese (C‐V) model is proposed for segmenting remote sensing images. Compared with the C‐V model, two main contributions of the proposed model mainly include the following: First, we introduce a new strategy of initialization in which the contours of the first k biggest connected regions are extracted as the initial curves (k is the number of level set functions); Second, to increase the accuracy, a morphological gradient component is added to the original intensity image. To investigate the effectiveness and efficiency of the proposed model, we have applied it to analyze different kinds of images, including synthetic, real, and remote sensing images. The experimental results have shown that our method is able to achieve better segmentation with less computational consumption compared with the traditional multiphase C‐V model and local and global intensity fitting model. Copyright © 2013 John Wiley & Sons, Ltd.

Evolving spiking neural networks (eSNN) are computational models that evolve new spiking neurons and new connections from incoming data to learn patterns from them in an on-line mode. With the development of new techniques to capture spatio- and spectro-temporal data in a fast on-line mode, using for example address event representation (AER) such as the implemented one in the artificial retina and the artificial cochlea chips, and with the available SNN hardware technologies, new and more efficient methods for spatio-temporal pattern recognition (STPR) are needed. The paper introduces a new eSNN model dynamic eSNN (deSNN), that utilises both rank-order spike coding (ROSC), also known as time to first spike, and temporal spike coding (TSC). Each of these representations are implemented through different learning mechanisms - RO learning, and temporal spike learning - spike driven synaptic plasticity (SDSP) rule. The deSNN model is demonstrated on a small scale moving object classification problem when AER data is collected with the use of an artificial retina camera. The new model is superior in terms of learning time and accuracy for learning. It makes use of the order of spikes input information which is explicitly present in the AER data, while a temporal spike learning rule accounts for any consecutive spikes arriving on the same synapse that represent temporal components in the learned spatio-temporal pattern.

In a previous work (Mohemmed et al. [11]), the authors proposed a supervised learning algorithm to train a spiking neuron to associate input/output spike patterns. In this paper, the association learning rule is applied in training a single layer of spiking neurons to classify multiclass spike patterns whereby the neurons are trained to recognize an input spike pattern by emitting a predetermined spike train. The training is performed in incremental fashion, i.e. the synaptic weights are adjusted after each presentation of a training pattern. The individual neurons are trained independently from other neurons and on patterns from a single class. A spike train comparison criterion is used to decode the output spike trains into class labels. The results of the simulation experiments on a synthetic dataset of spike patterns show a high efficiency in solving the considered classification task.

Spiking Neural Networks (SNN) were shown to be suitable tools for the processing of spatio-temporal information. However, due to their inherent complexity, the formulation of efficient supervised learning algorithms for SNN is difficult and remains an important problem in the research area. This article presents SPAN - a spiking neuron that is able to learn associations of arbitrary spike trains in a supervised fashion allowing the processing of spatio-temporal information encoded in the precise timing of spikes. The idea of the proposed algorithm is to transform spike trains during the learning phase into analog signals so that common mathematical operations can be performed on them. Using this conversion, it is possible to apply the well-known Widrow-Hoff rule directly to the transformed spike trains in order to adjust the synaptic weights and to achieve a desired input/output spike behavior of the neuron. In the presented experimental analysis, the proposed learning algorithm is evaluated regarding its learning capabilities, its memory capacity, its robustness to noisy stimuli and its classification performance. Differences and similarities of SPAN regarding two related algorithms, ReSuMe and Chronotron, are discussed.

To adapt linear discriminant analysis (LDA) to real-world applications, there is a pressing need to equip it with an incremental learning ability to integrate knowledge presented by one-pass data streams, a functionality to join multiple LDA models to make the knowledge sharing between independent learning agents more efficient, and a forgetting functionality to avoid reconstruction of the overall discriminant eigenspace caused by some irregular changes. To this end, we introduce two adaptive LDA learning methods: LDA merging and LDA splitting. These provide the benefits of ability of online learning with one-pass data streams, retained class separability identical to the batch learning method, high efficiency for knowledge sharing due to condensed knowledge representation by the eigenspace model, and more preferable time and storage costs than traditional approaches under common application conditions. These properties are validated by experiments on a benchmark face image data set. By a case study on the application of the proposed method to multiagent cooperative learning and system alternation of a face recognition system, we further clarified the adaptability of the proposed methods to complex dynamic learning tasks.

This paper proposes a new model of an Evolving Spiking Neural Network (ESNN) for spatio-temporal data (STD) classification problems. The proposed ESNN model incorporates an additional layer for capturing both spatial and temporal components of the STD and then transforms them into high dimensional spiking patterns. These patterns are learned and classified in the evolving classification layer of the ESNN. A fast time-to-first-spike learning algorithm is used that enables the new model to be more suitable for learning from the STD streams in an adaptive and incremental manner. The proposed method is evaluated on a benchmark sign language video that is spatio-temporal in nature. The results show that the proposed method is able to capture important spatio-temporal information from the STD stream. This results in significantly higher classification accuracy than the traditional time-delay MLP neural network model. Future directions for the development of ESNN models for STD are discussed.

We present a method that is based on Particle Swarm Optimization (PSO) for training a Spiking Neural Network (SNN) with dynamic synapses to generate precise time spike sequences. The similarity between the desired spike sequence and the actual output sequence is measured by a simple leaky integrate and fire spiking neuron. This measurement is used as a fitness function for PSO algorithm to tune the dynamic synapses until a desired spike output sequence is obtained when certain input spike sequence is presented. Simulations are made to illustrate the performance of the proposed method.

This study employs networks of stochastic spiking neurons as reservoirs for liquid state machines (LSM). We experimentally investigate the separation property of these reservoirs and show their ability to generalize classes of input signals. Similar to traditional LSM, probabilistic LSM (pLSM) have the separation property enabling them to distinguish between different classes of input stimuli. Furthermore, our results indicate some potential advantages of non-deterministic LSM by improving upon the separation ability of the liquid. Three non-deterministic neural models are considered and for each of them several parameter configurations are explored. We demonstrate some of the characteristics of pLSM and compare them to their deterministic counterparts. pLSM offer more flexibility due to the probabilistic parameters resulting in a better performance for some values of these parameters.

Time-series modeling and prediction have been very well researched by both the statistical and data mining communities. However, the multiple time-series problem of modeling and predicting simultaneous movements of a collection of time-sensitive variables that are related to each other has received much less attention. Strong relationships between variables suggest that trajectories of given variables involved in the relationships can be improved by including the nature and strength of these relationships in a prediction model. The key challenge is to capture the dynamics of the relationships to reflect changes that take place continuously over time. This research presents a global model to capture inclusive patterns of dynamic interactions between multiple time-series and a local trend model to extract localized profiles of relationships and recurring trends in multiple time-series. Our experimentation revealed that the global and local models specially developed for multiple time-series prediction outperformed methods such as multiple linear regression and the multilayer perceptron that were developed for predicting single time-series.

To adapt linear discriminant analysis (LDA) to real world applications, there is a pressing necessity to provide it with an incremental learning ability to integrate knowledge presented by one-pass data streams, a functionality to join multiple LDA models to make the knowledge-sharing between independent learning agents more efficient, and a forgetting functionality to avoid reconstruction of the overall discriminant eigenspace caused by some irregular changes. To this end, we introduce two adaptive LDA learning methods: LDA merging and LDA splitting, which show the following merits: ability of online learning with one-pass data streams, retained class separability identical to the batch learning method, high efficiency for knowledge-sharing due to condensed knowledge representation by the eigenspace model, and more preferable time and storage cost than traditional approaches under common application conditions. These properties are validated by the experiments on a benchmark face image dataset. By the case study on application of the proposed method to multi-agent cooperative learning and system alternation of a face recognition system, we further clarified the adaptability of the proposed methods to complex dynamic learning tasks.

This study investigates the characteristics of the Quantum-inspired Spiking Neural Network (QiSNN) feature selection and classification framework. The self-adapting nature of QiSNN due to the simultaneous optimization of network parameters and feature subsets represents a highly desirable characteristic in the context of machine learning and knowledge discovery. In this paper, the evolution of the parameters and feature subsets is studied in detail. The goal of this analysis is a comprehensive understanding of all parameters involved in QiSNN and some practical guidelines for using the method in future research and applications. We also highlight the role of the employed neural encoding technique along with its impact on the classification abilities of QiSNN.

The construction of a Spiking Neural Network (SNN), i.e. the choice of an appropriate topology and the configuration of its internal parameters, represents a great challenge for SNN based applications. Evolutionary Algorithms (EAs) offer an elegant solution for these challenges and methods capable of exploring both types of search spaces simultaneously appear to be the most promising ones. A variety of such heterogeneous optimization algorithms have emerged recently, in particular in the field of probabilistic optimization. In this paper, a literature review on heterogeneous optimization algorithms is presented and an example of probabilistic optimization of SNN is discussed in detail. The paper provides an experimental analysis of a novel Heterogeneous Multi-Model Estimation of Distribution Algorithm (hMM-EDA). First, practical guidelines for configuring the method are derived and then the performance of hMM-EDA is compared to state-of-the-art optimization algorithms. Results show hMM-EDA as a light-weight, fast and reliable optimization method that requires the configuration of only very few parameters. Its performance on a synthetic heterogeneous benchmark problem is highly competitive and suggests its suitability for the optimization of SNN.

This paper demonstrates how knowledge can be extracted from evolving spiking neural networks with rank order population coding. Knowledge discovery is a very important feature of intelligent systems. Yet, a disproportionally small amount of research is centered on the issue of knowledge extraction from spiking neural networks which are considered to be the third generation of artificial neural networks. The lack of knowledge representation compatibility is becoming a major detriment to end users of these networks. We show that a high-level knowledge can be obtained from evolving spiking neural networks. More specifically, we propose a method for fuzzy rule extraction from an evolving spiking network with rank order population coding. The proposed method was used for knowledge discovery on two benchmark taste recognition problems where the knowledge learnt by an evolving spiking neural network was extracted in the form of zero-order Takagi-Sugeno fuzzy IF-THEN rules.

In this paper, we propose a new Chunk IPCA algorithm in which an optimal threshold of accumulation ratio is adaptively selected such that the classification accuracy is maximized for a validation data set. In order to obtain a proper set of validation data, an online clustering method calledEvolvingClusteringMethod(ECM) is introduced into Chunk IPCA. In the proposed Chunk IPCA calledCIPCA-ECM, training data are first separated into the subsets of every class; then, ECM is applied to each subset to update the validation data set. In the experiments, the evaluation of the proposed Chunk IPCA algorithm is carried out using the four UCI data sets and the effectiveness of updating the threshold is discussed. The results suggest that the incremental learning of an eigenspace in the proposed CIPCA-ECM is stably carried out, and a compact and effective eigenspace is obtained over the entire learning stages. The recognition accuracy of CIPCA-ECM is almost equal to the best performance of CIPCA-FIX in which an optimal threshold is manually predetermined.

This paper presented a novel active linear discriminant analysis (LDA) learning method in the form of curiosity-driven incremental LDA (cILDA) and multiple cILDA agents cooperative learning (mcILDA). The curiosity in psychology here is modelled mathematically as a discriminability residue in-between instance space and its corresponding eigenspace. As the learning proceeds, the curiosity of an individual agent updates over time by two incremental learning processes: One updates the characterization of eigenspace and another re-calculates the curiosity. In the multi-agent scenario, individual agent communicates and cooperates with each other at every learning stage to discover the discriminant characterization of the whole pattern. In the experiment, we described how the discriminative instances could be significantly selected based on the curiosity with, at most, minor sacrifices in learning rate and classification accuracy. The experimental results show that the proposed curiosity learning performs gracefully under different level of redundancy, and the proposed cILDA/mcILDA learning system is capable of learning less instances, but has more often an improved discrimination performance.

The paper introduces a framework and implementation of an integrated connectionist system, where the features and the parameters of an evolving spiking neural network are optimised together using a quantum representation of the features and a quantum inspired evolutionary algorithm for optimisation. The proposed model is applied on ecological data modeling problem demonstrating a significantly better classification accuracy than traditional neural network approaches and a more appropriate feature subset selected from a larger initial number of features. Results are compared to a naive Bayesian classifier.

The quantum-inspired evolutionary algorithm (QEA) applies several quantum computing principles to solve optimization problems. In QEA, a population of probabilistic models of promising solutions is used to guide further exploration of the search space. This paper clearly establishes that QEA is an original algorithm that belongs to the class of estimation of distribution algorithms (EDAs), while the common points and specifics of QEA compared to other EDAs are highlighted. The behavior of a versatile QEA relatively to three classical EDAs is extensively studied and comparatively good results are reported in terms of loss of diversity, scalability, solution quality, and robustness to fitness noise. To better understand QEA, two main advantages of the multimodel approach are analyzed in details. First, it is shown that QEA can dynamically adapt the learning speed leading to a smooth and robust convergence behavior. Second, we demonstrate that QEA manipulates more complex distributions of solutions than with a single model approach leading to more efficient optimization of problems with interacting variables.

This paper introduces a novel neuro-fuzzy inference system denoted as ldquoMUFIS: a neuro-fuzzy inference system using multiple types of fuzzy rulesrdquo, for allowing multiple types of fuzzy rules to be used together to achieve a better performance. At each data point, the output of MUFIS is calculated through a fuzzy inference system based on m-most activated fuzzy rules which are dynamically chosen from multi-type fuzzy rules. It is demonstrated that MUFIS can effectively implement prediction and function approximation. We evaluate its performance on two case studies - a benchmark time-series prediction problem - Mackey Glass, and a real life medical prediction problem - glomerular filtration rate prediction.

This paper presents a novel method of rule extraction by encoding the knowledge of the data into an SVM classification tree (SVMT), and decoding the trained SVMT into a set of linguistic association rules. The method of rule extraction over the SVMT (r-SVMT), in the spirit of decision-tree rule extraction, achieves rule extraction not only from SVM, but also over the obtained decision-tree structure. The benefits of r-SVMT are that the decision-tree rule provides better comprehensibility, and the support-vector rule retains the good classification accuracy of SVM. Furthermore, the r-SVMT is capable of performing a very robust classification on such datasets that have seriously, even overwhelmingly, class-imbalanced data distribution, which profits from the super generalization ability of SVMT owing to the aggregation of a group of SVMs. Experiments with a gaussian synthetic data, seven benchmark cancers diagnosis have highlighted the utility of SVMT and r-SVMT on encoding and decoding rule knowledge, as well as the superior properties of r-SVMT as compared to a completely support-vector based rule extraction.

The paper investigates the use of the spiking neural networks for taste recognition in a simple artificial gustatory model. We present an approach based on simple integrate-and-fire neurons with rank order coded inputs where the network is built by an evolving learning algorithm. Further, we investigate how the information encoding in a population of neurons influences the performance of the networks. The approach is tested on two real-world datasets where the effectiveness of the population coding and networkpsilas adaptive properties are explored.

Most biological processes that are currently being researched in bioinformatics are complex, dynamic processes that are difficult to model and understand. The paper presents evolving connectionist systems (ECOS) as a general approach to adaptive modeling and knowledge discovery in bioinformatics. This approach extends the traditional machine learning approaches with various adaptive learning and rule extraction procedures. ECOS belong to the class of incremental local learning and knowledge-based neural networks. They are applied here to challenging problems in Bioinformatics, such as: microarray gene expression profiling, gene regulatory network (GRN) modeling, computational neurogenetic modeling. The ECOS models have several advantages when compared to the traditional techniques: fast learning, incremental adaptation to new data, facilitating knowledge discovery through fuzzy rules. © 2008 Wiley Periodicals, Inc.

It is a well-recognized fact that the theory of fuzzy sets and systems, for the last four decades after the seminal paper by Professor Zadeh [1], has demonstrated its remarkable ability to go beyond conventional information representation. It resulted in a wide range of new formulations of practical problems, such as fuzzy control, fuzzy clustering and classification, fuzzy modeling, and fuzzy optimization [2]. Historically, the design of the fuzzy systems has been initially assumed to be centered on expert knowledge [3]. During the 1990s, a new trend emerged [4], [5] that offered techniques to make use of the experimental data. This data-centered approach can be used to enhance and validate the existing expert knowledge or can also be used to substitute its lack (as is the case with autonomous systems, for example). Neurofuzzy and hybrid learning systems were introduced, where fuzzy representation was integrated into a neural learning architecture to bring linguistic meaning of the learned information [5]. (c) IEEE Press

This paper presents a pattern classification system in which feature extraction and classifier learning are simultaneously carried out not only online but also in one pass where training samples are presented only once. For this purpose, we have extended incremental principal component analysis (IPCA) and some classifier models were effectively combined with it. However, there was a drawback in this approach that training samples must be learned one by one due to the limitation of IPCA. To overcome this problem, we propose another extension of IPCA called chunk IPCA in which a chunk of training samples is processed at a time. In the experiments, we evaluate the classification performance for several large-scale data sets to discuss the scalability of chunk IPCA under one-pass incremental learning environments. The experimental results suggest that chunk IPCA can reduce the training time effectively as compared with IPCA unless the number of input attributes is too large. We study the influence of the size of initial training data and the size of given chunk data on classification accuracy and learning time. We also show that chunk IPCA can obtain major eigenvectors with fairly good approximation.

This study points out some weaknesses of existing quantum-inspired evolutionary algorithms (QEA) and explains in particular how hitchhiking phenomena can slow down the discovery of optimal solutions and encourage premature convergence. A new algorithm, called versatile quantum- inspired evolutionary algorithm (vQEA), is proposed. With vQEA, the attractors moving the population through the search space are replaced at every generation without considering their fitness. The new algorithm is much more reactive. It always adapts the search toward the last promising solution found thus leading to a smoother and more efficient exploration. In this paper, vQEA is tested and compared to a classical genetic algorithm CGA and to a QEA on several benchmark problems. Experiments have shown that vQEA performs better than both CGA and QEA in terms of speed and accuracy. It is a highly scalable algorithm as well. Finally, the properties of the vQEA are discussed and compared to estimation of distribution algorithms (EDA).

Evolving Connectionist Systems (ECOS) are neural network systems that develop their structure, functionality and internal representation through continuous learning from data and interaction with the environment. ECOS can also evolve through generations of populations using evolutionary computation, but the focus of the presentation is on: (1) Adaptive learning and improvement of each individual model; (2) Knowledge representation, knowledge adaptation and knowledge extraction. The learning process can be: on-line, off-line, incremental, supervised, unsupervised, active, sleep/dream, etc.

This article reports on our brain-gene ontology (BGO) system that we use as a tool for educational purpose and research. We present some preliminary results on the brain-gene ontology (BGO) project that is concerned with the collection, presentation and use of knowledge in the form of ontology. BGO includes various concepts, facts, data, software simulators, graphs, animations, and other information forms, related to brain functions, brain diseases, their genetic basis and the relationship between all of them. The first version of the brain-gene ontology has been completed as a hierarchical structure and as an initial implementation in the Protege ontology building environment.

This article presents some preliminary results on a brain-gene ontology project that is concerned with the collection and the presentation, in a form of ontology, of various concepts, facts, data, software simulators, graphs, videos, animations, and other information forms, related to brain functions, brain diseases, their genetic basis and the relationship between all of them. The first version of the brain-gene ontology (BGO) has been completed as a structure and as an initial implementation in the Protégé ontology-building environment. The BGO allows users to: navigate through the rich information space of brain functions and brain diseases, brain related genes and their activities in certain parts of the brain and their relation to brain diseases; to run simulations; to download data that can be used in a software machine learning environment such as WEKA and NeuCom to train prediction or classification models; to visualize relationship information; to add some new information as the BGO has an evolving structure. The BGO is designed to facilitate active learning and research in the areas of bioinformatics, neuroinformatics, information engineering, and knowledge management. Different parts of it can be used by different users, from a school level to postgraduate and PhD student level. A further development of the BGO is discussed, where more data and information will be added, that will include both a higher level information on cognitive functions and consciousness, and a lower level quantum information.

This paper presents a new algorithm of dynamic feature selection by extending the algorithm of Incremental Principal Component Analysis (IPCA), which has been originally proposed by Hall and Martin. In the proposed IPCA, a chunk of training samples can be processed at a time to update the eigenspace of a classification model without keeping all the training samples given so far. Under the assumption that L of training samples are given in a chunk, first we derive a new eigenproblem whose solution gives us a rotation matrix of eigen-axes, then we introduce a new algorithm of augmenting eigen-axes based on the accumulation ratio. We also derive the one-pass incremental update formula for the accumulation ratio. The experiments are carried out to verify if the proposed IPCA works well. Our experimental results demonstrate that it works well independent of the size of data chunk, and that the eigenvectors for major components are obtained without serious approximation errors at the final learning stage. In addition, it is shown that the proposed IPCA can maintain the designated accumulation ratio by augmenting new eigen-axes properly. This property enables a learning system to construct an informative eigenspace with minimum dimensionality.

We present new results from computational neurogenetic modeling to aid discoveries of complex gene interactions underlying oscillations in neural systems. Interactions of genes in neurons affect the dynamics of the whole neural network model through neuronal parameters, which change their values as a function of gene expression. Through optimization of the gene interaction network, initial gene/protein expression values and neuronal parameters, particular target states of the neural network operation can be achieved, and statistics about gene interaction matrix can be extracted. In such a way it is possible to model the role of genes and their interactions in different brain states and conditions. Experiments with human EEG data are presented as an illustration of this methodology and also, as a source for the discovery of unknown interactions between genes in relation to their impact on brain activity.

The paper presents a methodology for using computational neurogenetic modelling (CNGM) to bring new original insights into how genes influence the dynamics of brain neural networks. CNGM is a novel computational approach to brain neural network modelling that integrates dynamic gene networks with artificial neural network model (ANN). Interaction of genes in neurons affects the dynamics of the whole ANN model through neuronal parameters, which are no longer constant but change as a function of gene expression. Through optimization of interactions within the internal gene regulatory network (GRN), initial gene/protein expression values and ANN parameters, particular target states of the neural network behaviour can be achieved, and statistics about gene interactions can be extracted. In such a way, we have obtained an abstract GRN that contains predictions about particular gene interactions in neurons for subunit genes of AMPA, GABAA and NMDA neuro-receptors. The extent of sequence conservation for 20 subunit proteins of all these receptors was analysed using standard bioinformatics multiple alignment procedures. We have observed abundance of conserved residues but the most interesting observation has been the consistent conservation of phenylalanine (F at position 269) and leucine (L at position 353) in all 20 proteins with no mutations. We hypothesise that these regions can be the basis for mutual interactions. Existing knowledge on evolutionary linkage of their protein families and analysis at molecular level indicate that the expression of these individual subunits should be coordinated, which provides the biological justification for our optimized GRN.

The paper presents a novel, biologically plausible spiking neuronal model that includes a dynamic gene network. Interactions of genes in neurons affect the dynamics of the neurons and the whole network through neuronal parameters that change as a function of gene expression. The proposed model is used to build a spiking neural network (SNN) illustrated on a real EEG data case study problem. The paper also presents a novel computational approach to brain neural network modeling that integrates dynamic gene networks with a neural network model. Interaction of genes in neurons affects the dynamics of the whole neural network through neuronal parameters, which are no longer constant, but change as a function of gene expression. Through optimization of the gene interaction network, initial gene/protein expression values and ANN parameters, particular target states of the neural network operation can be achieved, and statistics about gene intercation matrix can be extracted. It is illustrated by means of a simple neurogenetic model of a spiking neural network (SNN). The behavior of SNN is evaluated by means of the local field potential, thus making it possible to attempt modeling the role of genes in different brain states, where EEG data is available to test the model. We use standard signal processing techniques like FFT to evaluate the SNN output to compare it with real human EEG data.

While inductive modeling is used to develop a model (function) from data of the whole problem space and then to recall it on new data, transductive modeling is concerned with the creation of single model for every new input vector based on some closest vectors from the existing problem space. The model approximates the output value only for this input vector. However, deciding on the appropriate distance measure, on the number of nearest neighbors and on a minimum set of important features/variables is a challenge and is usually based on prior knowledge or exhaustive trial and test experiments. This paper proposes a genetic algorithm (GA) approach for optimizing these three factors. The method is tested on several datasets from UCI repository for classification tasks and results show that it outperforms conventional approaches. The drawback of this approach is the computational time complexity due to the presence of GA, which can be overcome using parallel computer systems due to the intrinsic parallel nature of the algorithm.

In this paper, a new approach to face recognition is presented in which not only a classifier but also a feature space of input variables is learned incrementally to adapt to incoming training samples. A benefit of this type of incremental learning is that the search for useful features and the learning of an optimal decision boundary are carried out in an online fashion. To implement this idea, an extended version of incremental principal component analysis (IPCA) and resource allocating network with long-term memory (RAN-LTM) are effectively combined. Using IPCA, a feature space is updated by rotating its eigen-axes and increasing the dimensions to adapt to a new training sample. In RAN-LTM, a small number of training samples called memory items are selected and they are utilized for retraining a classifier to realize an excellent incremental ability. To accommodate the classifier to the evolution of the feature space, we present a way to reconstruct the neural classifier without keeping all of the training samples given previously. In the experiments, the proposed incremental learning model is evaluated over a self-compiled face image database. As the result, we verify that the proposed model works well without serious forgetting and the test performance is improved as the learning stages proceed.

This paper introduces a novel neural fuzzy inference method-NFI for transductive reasoning systems. NFI develops further some ideas from DENFIS-dynamic neuro-fuzzy inference systems for both online and offline time series prediction tasks. While inductive reasoning is concerned with the development of a model (a function) to approximate data in the whole problem space (induction), and consecutively-using this model to predict output values for a new input vector (deduction), in transductive reasoning systems a local model is developed for every new input vector, based on some closest to this vector data from an existing database (also generated from an existing model). NFI is compared with both inductive connectionist systems (e.g., MLP, DENFIS) and transductive reasoning systems (e.g., K-NN) on three case study prediction/identification problems. The first one is a prediction task on Mackey Glass time series; the second one is a classification on Iris data; and the last one is a real medical decision support problem of estimating the level of renal function of a patient, based on measured clinical parameters for the purpose of their personalised treatment. The case studies have demonstrated better accuracy obtained with the use of the NFI transductive reasoning in comparison with the inductive reasoning systems.

This letter proposes a new negative correlation (NC) learning method that is both easy to implement and has the advantages that: 1) it requires much lesser communication overhead than the standard NC method and 2) it is applicable to ensembles of heterogenous networks.

This paper presents a constructive method for deriving an updated discriminant eigenspace for classification when bursts of data that contains new classes is being added to an initial discriminant eigenspace in the form of random chunks. Basically, we propose an incremental linear discriminant analysis (ILDA) in its two forms: a sequential ILDA and a Chunk ILDA. In experiments, we have tested ILDA using datasets with a small number of classes and small-dimensional features, as well as datasets with a large number of classes and large-dimensional features. We have compared the proposed ILDA against the traditional batch LDA in terms of discriminability, execution time and memory usage with the increasing volume of data addition. The results show that the proposed ILDA can effectively evolve a discriminant eigenspace over a fast and large data stream, and extract features with superior discriminability in classification, when compared with other methods.

Many systems in Biology and Nature are characterized by a continuous adaptation and by a complex interaction of many variables over time.Such systems can be observed at different levels of the functioning of a living organism, e.g.: molecular, genetic, cellular, multicellular, neuronal, brain function, evolution.One of the challenges for information science is to be able to represent the dynamic processes, to model them, and to reveal "the rules" that govern the adaptation and the variable interaction over time.

The paper introduces a combination of adaptive neural network systems and statistical method for integrating speech and face image information for person identification. The method allows for the development of models of persons and their on-going adjustment based on new speech and face images. The method is illustrated with a modeling and classification of different persons, when speech and face images are presented in an incremental way. In this model, there are two subnetworks, one for face image and one for speaker recognition. A higher-level layer is applied to make a final decision. In the speaker recognition subnetwork, a text-dependant model is built using evolving connectionist systems (ECOS) [N. Kasabov, 2002]. In the face image recognition subnetwork, composite profile technique is applied for face image feature extraction and zero instruction set computing (ZISC) [ZISC Manual, 2000] technology is used to build the neural network. In the higher-level conceptual subsystem, final recognition decision is made using statistical method. The experiments show that ECOS and ZISC are appropriate techniques for the creation of evolving models for the task of speaker and face recognition individually. It is also shown that the integration of the speech and image information using statistical method improves the person identification rate.

Clustering time course gene expression data (gene trajectories) is an important step towards solving the complex problem of gene regulatory network (GRN) modeling and discovery as it significantly reduces the dimensionality of the gene space required for analysis. This paper introduces a novel method that hybridizes genetic algorithm (GA) and expectation maximization algorithms (EM) for clustering with the mixtures of multiple linear regression models (MLRs). The proposed method is applied to cluster gene expression time course data into smaller number of classes based on their trajectory similarities. Its performance and application as a generic clustering method to other complex problems are discussed.

This paper introduces a novel connectionist approach to neural network modelling that integrates dynamic gene networks within neurons with a neural network model. Interaction of genes in neurons affects the dynamics of the whole neural network. Through tuning the gene interaction network and the initial gene/protein expression values, different states of the neural network operation can be achieved. A generic computational neurogenetic model is introduced that implements this approach. It is illustrated by means of a simple neurogenetic model of a spiking neural network (SNN). Functioning of the SNN can be evaluated for instance by the field potentials, thus making it possible to attempt modelling the role of genes in different brain states such as epilepsy, schizophrenia, and other states, where EEG data is available to test the model predictions.

This paper compares inductive-, versus transductive modeling, and also global-, versus local models with the use of SVM for gene expression classification problems. SVM are used in their three variants - inductive SVM, transductive SVM (TSVM), and SVM tree (SVMT) - the last two techniques being recently introduced by the authors. The problem of gene expression classification is used for illustration and four benchmark data sets are used to compare the different SVM methods. The TSVM outperforms the inductive SVM models applied on a small to medium variable (gene) set and a small to medium sample set, while SVMT is superior when the problem is defined with a large data set, or - a large set of variables (e.g. 7,000 genes, with little or no variable pre-selection).

Abs.

This work applies Evolutionary Computation to achieve completely self-adapting Evolving Fuzzy Neural Networks (EFuNNs) for operating in both incremental (on-line) and batch (off-line) modes. EFuNNs belong to a class of Evolving Connectionist Systems (ECOS), capable of performing clustering-based, on-line, local area learning and rule extraction. Through Evolutionary Computation, its parameters such as learning rates and membership functions are continuously adjusted to reflect the changes in the dynamics of incoming data. The proposed methods are tested on the Mackey–Glass series and the results demonstrate a substantial improvement in EFuNN's performance.

Research with microarray gene expression analysis has primarily been on expression profiling based on one set of microarray data. This paper presents a novel approach to integrated analysis and modeling of microarray data from multiple sources. Normalization method is applied to different data sets before they are used together in an adaptive connectionist classification system. The method is demonstrated on a bench-mark case study problem of classifying Diffuse Large B-cell lymphoma (DLBCL) and Follicular lymphoma (FL). For the purpose of comparison, different normalization techniques were applied and connectionist models were created from one or more microarray data sets and then tested on the others. The results show that with the use of proper normalization and modeling techniques, a model based on one set of data can be used to classify microarray data from totally different sources. For the modeling part, evolving connectionist systems (ECOS) are used that allow for new data to be added in an incremental way so that connectionist systems can be built for on-line adaptive learning where new data from various sources can be added into the system.

The paper describes a method of using evolutionary computation technique for parameter optimisation of evolving connectionist systems (ECOS) that operate in an online, life-long learning mode. ECOS evolve their structure and functionality from an incoming stream of data in either a supervised-, of/and in an unsupervised mode. The algorithm is illustrated on a case study of predicting a chaotic time-series that changes its dynamics over time. With the on-line parameter optimisation of ECOS, a faster adaptation and a better prediction is achieved. The method is practically applicable for real time applications.

The inverse problem of recovering the potential transformer primary signal waveform using secondary signal waveform and information about the secondary load is solved here via two inverse neural network models. The first model uses two recurrent neural networks trained in an off-line mode. The second model is designed with the use a dynamic evolving neural-fuzzy interface system (DENFIS) and suited for online application and integration into existing protection algorithms as a parallel module. It has the ability of learning and adjusting its structure in an online mode to reflect changes in the environment. The model is suited for real time applications and improvement of protection relay operation. The two models perform better than any existing and published models so far and are useful not only for the reconstruction of the primary signal, but for predicting the signal waveform for some time steps ahead and thus for estimating the drifts in the incoming signals and events.

The paper presents a method for optimising parameter values of evolving connectionist systems (ECoS) for life-long learning. The method is based on evolutionary computation principles, and on genetic algorithms in particular. The method is illustrated on a spoken phoneme data classification task.

Microarray techniques have recently made it possible to monitor simultaneously the activity of thousands of genes. They offer new insights into the biology of a cell. However, the data produced by microarrays poses several challenges to overcome. One major task in the analysis of microarray data is to reveal structures in the data despite its large noise component. We used fuzzy c-means (FCM) clustering in this study to achieve a robust analysis of gene expression time-series. We address the issues of parameter selection and cluster validity. Using statistical models to simulate gene expression data, we show that FCM can detect genes belonging to different classes. This may open the way for the study of fine-structures in microarray data.

This paper introduces a new type of fuzzy inference systems, denoted as dynamic evolving neural-fuzzy inference system (DENFIS), for adaptive online and offline learning, and their application for dynamic time series prediction. DENFIS evolve through incremental, hybrid (supervised/unsupervised), learning, and accommodate new input data, including new features, new classes, etc., through local element tuning. New fuzzy rules are created and updated during the operation of the system. At each time moment, the output of DENFIS is calculated through a fuzzy inference system based on m-most activated fuzzy rules which are dynamically chosen from a fuzzy rule set. Two approaches are proposed: (1) dynamic creation of a first-order Takagi-Sugeno-type fuzzy rule set for a DENFIS online model; and (2) creation of a first-order Takagi-Sugeno-type fuzzy rule set, or an expanded high-order one, for a DENFIS offline model. A set of fuzzy rules can be inserted into DENFIS before or during its learning process. Fuzzy rules can also be extracted during or after the learning process. An evolving clustering method (ECM), which is employed in both online and offline DENFIS models, is also introduced. It is demonstrated that DENFIS can effectively learn complex temporal sequences in an adaptive way and outperform some well-known, existing models.

The paper presents a method based on evolution strategies that attempts to optimise the training parameters of a class of online, adaptive connectionist-based learning systems called evolving connectionist systems (ECoS). ECoS are systems that evolve their structure and functionality through online, adaptive learning from incoming data. The ECoS paradigm is combined with the paradigm of evolutionary computation to attempt to solve a difficult task of online adaptive adjustment and optimisation of the parameter values of the evolving system. Although the method presented is unsuccessful, some useful information about the properties of the ECoS model is still derived from the work.

This paper introduces an extension to the existing theory of the evolving fuzzy neural network (EFuNN) for it to be a multi-module classifier as well. We call this proposed architecture multi-EFuNN. The incorporation of the evolving clustering method is used to partition the input space of the dataset and also determine how many EFuNNs are to be used to classify it. The main advantages of this multi-module classifier is in the areas of online learning and recall of data where there are a growing number of classes with more data coming. Preliminary results conducted using this architecture are compared to the existing single EFuNN classifier and reported.

Although widely studied for many years, colour quantisation remains a practical problem in image processing. Unlike previous works where the image can only be quantised after the whole set of image data is acquired, we propose to use an evolving localised learning model for on-line colour quantisation. This approach is compared with some conventional algorithms.

This paper introduces evolving fuzzy neural networks (EFuNNs) as a means for the implementation of the evolving connectionist systems (ECOS) paradigm that is aimed at building online, adaptive intelligent systems that have both their structure and functionality evolving in time. EFuNNs evolve their structure and parameter values through incremental, hybrid supervised/unsupervised, online learning. They can accommodate new input data, including new features, new classes, etc., through local element tuning. New connections and new neurons are created during the operation of the system. EFuNNs can learn spatial-temporal sequences in an adaptive way through one pass learning and automatically adapt their parameter values as they operate. Fuzzy or crisp rules can be inserted and extracted at any time of the EFuNN operation. The characteristics of EFuNNs are illustrated on several case study data sets for time series prediction and spoken word classification. Their performance is compared with traditional connectionist methods and systems. The applicability of EFuNNs as general purpose online learning machines, what concerns systems that learn from large databases, life-long learning systems, and online adaptive systems in different areas of engineering are discussed.

Reports an application of an evolving fuzzy neural network (EFuNN) for camera operations recognition. EFuNN features one-pass learning, dynamical growing and shrinking architecture and ability to accommodate new knowledge without the need to retrain the network on both the original and new data. The network learns from pre-classified examples in the form of motion vector patterns, extracted from MPEG compressed video, in order to distinguish between six classes: static, panning, zooming, object motion, tracking and dissolve. The performance of EFuNN is compared with LVQ and the results are discussed. In addition, the impact of the number of membership functions and the contribution of the rule node aggregation are analyzed.

An algorithm of evolving self-organizing map (ESOM) is proposed as a dynamic version of the Kohonen self-organizing map, where network structure is evolved in an online adaptive mode. Experiments have been carried out on some benchmark data sets as well as on macroeconomic data. Results show that ESOM is a good tool for clustering, data analysis, and visualisation.

Speech and signal processing technologies need new methods that deal with the problems of noise and adaptation in order for these technologies to become common tools for communication and information processing. This paper is concerned with a method and a system for adaptive speech recognition in a noisy environment (ASN). A system based on the described method can store words and phrases spoken by the user and subsequently recognize them when they are pronounced as connected words in a noisy environment. The method guarantees system robustness in respect to noise, regardless of its origin and level. New words, pronunciations, and languages can be introduced to the system in an incremental, adaptive mode. The method and system are based on novel techniques recently created by the authors, namely: adaptive noise suppression, and evolving connectionist systems. Potential applications are numerous, e.g. voice dialling in a noisy environment, voice command control, improved wireless communications, data entry into databases, helping disabled people, multimedia systems, improved human computer interaction. The method and system are illustrated on the recognition of English and Italian spoken digits in different noisy environments.

The paper introduces evolving connectionist systems (ECOS) as an effective approach to building online adaptive intelligent systems. ECOS evolve through incremental, hybrid (supervised/unsupervised), online learning. They can accommodate new input data, including new features, new classes, etc. through local element tuning. New connections and new neurons are created during the operation of the system. The ECOS framework is presented and illustrated on a particular type of evolving neural networks-evolving fuzzy neural network (EFuNN). EFuNN can learn spatial-temporal sequences in an adaptive way, through one pass learning. Rules can be inserted and extracted at any time of the system operation. The characteristics of ECOS and EFuNN are illustrated on a case study of adaptive, phoneme-based spoken language recognition.

The AVIS framework for integrated audio and visual information processing is applied to the problem of person identification. An instantiation of the AVIS framework, called PIAVI, is based on a fuzzy neural network (FuNN) model of audio-visual person identification. In PIAVI's unimodal (visual) mode of operation, only dynamic visual features are used, whereas in the bimodal mode of operation, dynamic auditory and dynamic visual features are integrated at an early level of processing. Using a new dataset of dynamic features, a comparative study is performed with PIAVI in its two modes of operation. The results show that, with a large training set, perfect performance is achieved in the unimodal case. With a smaller training set, online application of the person identification system becomes feasible. Using this smaller set, unimodal identification performance is unsatisfactory. However, in the bimodal case, the identification performance is upgraded to satisfactory level of performance by early integration. It is concluded that, by using dynamic audio-visual features and FuNNs, an adequate on-line application of PIAVI in person identification tasks is within reach.

This paper explores a multimodular architecture of an intelligent information system and proposes a method for adaptation. The method is based on evaluating which of the modules need to be adapted based on the performance of the whole system on new data. These modules are then trained selectively on the new data until they improve their performance and the performance of the whole system. The modules are fuzzy neural networks, especially designed to facilitate adaptive training and knowledge discovery, and spatial temporal maps. A particular case study of spoken language recognition is presented along with some preliminary experimental results of an adaptive speech recognition system. © 1998 John Wiley & Sons, Inc.

A connectionist-based time-series analysis method is described that includes chaotic characterization, fractal analysis together with statistical data processing in an adaptive fuzzy neural network environment. The applied fuzzy neural network (FuNN) can utilize as well as generate knowledge during an iterative learning and adaptation procedure. Two major aspects of the present work are (1) incorporating knowledge into the fuzzy neural network based on the nonlinear deterministic, chaotic analysis of the signals and (2) refining and updating the knowledge base by the FuNN using adaptive learning techniques. Examples include the standard gas-furnace benchmark data analysis and also an application to a case study of multivariate signal analysis as part of a project for establishing a plantwise monitoring and process control system. © 1998 John Wiley & Sons, Inc.