Temporal Pattern Classification Research Articles

Classification of Spatial Temporal Patterns Based on Neuromorphic Networks

This work is devoted to the problems of developing neuromorphic classifiers of spatiotemporal patterns, as well as their application in neurointerfaces. Classifiers of spatiotemporal patterns based on neural networks, support vector machines, deep neural networks, and Riemannian geometry are considered. A comparative study of these classifiers is carried out in the plane of the accuracy of multiclass recognition of electroencephalographic signals showing time-dependent bioelectrical activity in different areas of the brain during the imagination of different movements. It is shown that such classifiers can provide an accuracy of 60-80% when recognizing from two to four classes of imaginary movements. A new type of classifier based on a neuromorphic network, based on the biosimilar neurons built on the Izhikevich model, is proposed. The network processes input spike sequences and generates pulse streams of different frequencies at the outputs. The network is trained using the Supervised STDP algorithm based on labeled information containing examples of the correct recognition of the required pattern classes. The recognized pattern class is determined by the maximum frequency of the output sequence. The neuromorphic classifier showed an average classification accuracy of 90% for 4 classes of imaginary commands and a maximum of 95%. By modeling the robot control task in the virtual environment it is shown that such accuracy is sufficient for the effective use of the classifier as part of a non-invasive brain-computer interface for non-contact control of robotic devices.

Real-Time Monitoring of Miniaturized Thermal Food Processing by Advanced Mass Spectrometric Techniques

Mass spectrometry is a popular and powerful analytical tool to study the effects of food processing. Industrial sampling, real-life sampling, or challenging academic research on process-related volatile and aerosol research often demand flexible, time-sensitive data acquisition by state-of-the-art mass analyzers. Here, we show a laboratory-scaled, miniaturized, and highly controllable setup for the online monitoring of aerosols and volatiles from thermal food processing based on dielectric barrier discharge ionization (DBDI) mass spectrometry (MS). We demonstrate the opportunities offered by the setup from a foodomics perspective to study emissions from the thermal processing of wheat bread rolls at 210 °C by Fourier transformation ion cyclotron resonance MS. As DBDI is an emerging technology, we compared its ionization selectivity to established atmospheric pressure ionization tools: we found DBDI preferably ionizes saturated, nitrogenous compounds. We likewise identified a sustainable overlap in the selectivity of detected analytes with APCI and electrospray ionization (ESI). Further, we dynamically recorded chemical fingerprints throughout the thermal process. Unsupervised classification of temporal response patterns was used to describe the dynamic nature of the reaction system. Compared to established tools for real-time MS, our setup permits one to monitor chemical changes during thermal food processing at ultrahigh resolution, establishing an advanced perspective for real-time mass spectrometric analysis of food processing.

A Simple Memristive Circuit for Pattern Classification Based on Reservoir Computing

Reservoir computing is a machine learning framework suited for pattern recognition of temporal signals. A reservoir computing system normally consists of a dynamic reservoir and a static readout. Memristive systems and devices with intrinsic nonlinearity are potentially favorable for constructing a reservoir hardware because the role of the reservoir is to nonlinearly map input signals into a higher-dimensional feature space. In this study, we propose a minimal configuration of memristive circuit reservoirs for solving temporal pattern classification problems. We numerically demonstrate how the classification performance of the proposed reservoir computing systems with only two memristive devices depends on the system conditions in waveform and electrocardiogram classification tasks. The signal processing methods used for performance improvement provide useful insights into effective and efficient implementation of memristive reservoir computing hardware.

Spatiotemporal Evolution and Regional Differences in the Production-Living-Ecological Space of the Urban Agglomeration in the Middle Reaches of the Yangtze River

The urban agglomeration in the middle reaches of the Yangtze River, which is the second largest urban agglomeration in China, represents a typical land space range of ecological vulnerability in China. Large differences occur in economic development mode between resource- and non-resource-based cities in this basin area. Accurate identification of the evolution and regional differences in the production-living-ecological space (PLES) is very important in order to elucidate the development and utilization of land space in the region. At present, relevant research has largely focused on the classification and determination of PLES temporal and spatial patterns. Temporal and spatial pattern research has mainly considered a single scale of administrative division, whereas fewer studies have analyzed the temporal and spatial patterns and regional differences in the PLES in ecologically fragile natural watersheds. Therefore, based on PLES classification, the regional differences in the PLES between two types of cities in the basin are measured via the Theil index and exploratory spatial data analysis (ESDA). First, the ecological space (ES) of these two types of cities in the urban agglomeration in the middle reaches of the Yangtze River is compressed by the production space (PS) and living space (LS), in which the ES of resource-based cities is compressed for a longer period, and the phenomenon involving PS compression by the LS and ES mainly occurs in non-resource-based cities within the urban agglomeration in the middle reaches of the Yangtze River. Second, the PLES of these two types of cities exhibits the characteristics of spatial aggregation, and high- and low-density areas of the PLES remain relatively stable. Third, the regional differences in the PLES of the urban agglomeration in the middle reaches of the Yangtze River mainly originate from intraregional differences. The PLES of these two types of cities in the urban agglomeration in the middle reaches of the Yangtze River is more sensitive to changes in economic development than to those in the population distribution.

Temporal Pattern Classification and Analysis of Large Volume of Data Using Associated Data Placement Algorithm and Online Community Adjustment

Aim: The main motto of the study is to optimize the large volume of data using data placement algorithm and online community adjustment algorithm and comparing their accuracy. Materials and Methods: Data placement algorithm (N=10) and online community adjustment (N=10) was iterated 20 times to optimize the data. Result and Discussion: Data placement algorithm has significantly better accuracy (85%) compared to online community adjustment algorithm(78%). The statistical significance of data placement (p<0.02 independent sample test) is high. Conclusion: With the limits of the study, a data placement algorithm with product manufacturing data offers the best accuracy in data optimization.

EVI Time-Series Breakpoint Detection Using Convolutional Networks for Online Deforestation Monitoring in Chaco Forest

The Dry Chaco Forest has the highest absolute deforestation rates of all Argentinian forests (current deforestation rate of 150 000 ha yr−1, 0.85% yr−1). The deforestation process is seen as a breakpoint in the enhanced vegetation index (EVI) time series, associated with the change from a typical forest phenology pattern to something else (e.g., bare soil, pasture, and cropland). Therefore, to monitor this process, a near real-time time-series breakpoint-detection model is needed. In this article, we exploited the 18-year-long MODIS EVI time-series data to train a temporal pattern classification model based on convolutional neural networks. Model architecture parameters (optimizer, number of hidden layers, number of neurons, and so on) were selected using an optimization procedure. The trained model then tries to estimate the probability that a given “time-series segment” corresponds to a deforestation event. The model was validated using in situ data derived from high-resolution images. Results are promising, since the model presents good performance for the validation data set [F1-score = 0.85, ${fpr} = 0.0012$ (of the order of the true deforestation rate), ${tpr} = 0.8$ , for a sample size = $50 \times 10^{3}$ ] and average performance in a yearly analysis (F1-score = 0.6, sample size = $1120 \times 10^{3}$ ). Model performance was studied using two diagnostic tools: activation maps and model ensemble error estimations. Results show that proposed model presents good extrapolation capabilities, but its maximum F1-score is bounded by error in the available data set (in particular, mislabeled deforestation events).

Reservoir Computing Using Diffusive Memristors

Reservoir computing (RC) is a framework that can extract features from a temporal input into a higher‐dimension feature space. The reservoir is followed by a readout layer that can analyze the extracted features to accomplish tasks such as inference and classification. RC systems inherently exhibit an advantage, since the training is only performed at the readout layer, and therefore they are able to compute complicated temporal data with a low training cost. Herein, a physical reservoir computing system using diffusive memristor‐based reservoir and drift memristor‐based readout layer is experimentally implemented. The rich nonlinear dynamic behavior exhibited by a diffusive memristor due to Ag migration and the robust in situ training of drift memristor arrays makes the combined system ideal for temporal pattern classification. It is then demonstrated experimentally that the RC system can successfully identify handwritten digits from the Modified National Institute of Standards and Technology (MNIST) dataset, achieving an accuracy of 83%.

Advancing the neurocomputer

Abstract Cultivating neural populations onto microelectrode arrays (MEAs) is a popular technique used by neuroscientists to study cortical microcircuits in a more controlled setting than they appear in vivo. However, recent works show that these neural cultures may be more than just another analog to the brain—they can also be used to compute. Researchers in academia and industry may soon need the computing power of a true neuromorphic computer: a silicon computing device that utilizes the processing power of biological neurons. Such a device can be realized by cultivating biological neurons onto an MEA in vitro and carefully interfacing the MEA to a computer or robot. This new device can make use of short- and long-term memory mechanisms intrinsic to biological neurons to train neuronal cultures to perform unique computations. Research shows this type of computing device may be able to solve problems in fields like image recognition, temporal pattern classification, and others. In this paper, we review the methods of interfacing cultures with MEAs, learning mechanisms exhibited in neuronal cultures, the current applications of a neuron–silicon hybrid computing device in research, and potential challenges this field may face. We propose that the current research in neurocomputers has provided the foundations for a new era in computing, one in which the computational power of true biological neural networks may be exploited and eventually surpass the power of artificial neural networks.

Digital Implementation of a Single Dynamical Node Reservoir Computer

Minimal hardware implementations of machine-learning techniques have been attracting increasing interest over the last decades. In particular, field-programmable gate array (FPGA) implementations of neural networks (NNs) are among the most appealing ones, given the match between system requirements and FPGA properties, namely, parallelism and adaptation. Here, we present an FPGA implementation of a conceptually simplified version of a recurrent NN based on a single dynamical node subject to delayed feedback. We show that this configuration is capable of successfully performing simple real-time temporal pattern classification and chaotic time-series prediction.

Automatic classification of acetowhite temporal patterns to identify precursor lesions of cervical cancer

Cervical cancer has remained, until now, as a serious public health problem in developing countries. The most common method of screening is the Pap test or cytology. When abnormalities are reported in the result, the patient is referred to a dysplasia clinic for colposcopy. During this test, a solution of acetic acid is applied, which produces a color change in the tissue and is known as acetowhitening phenomenon. This reaction aims to obtaining a sample of tissue and its histological analysis let to establish a final diagnosis. During the colposcopy test, digital images can be acquired to analyze the behavior of the acetowhitening reaction from a temporal approach. In this way, we try to identify precursor lesions of cervical cancer through a process of automatic classification of acetowhite temporal patterns. In this paper, we present the performance analysis of three classification methods: kNN, Naïve Bayes and C4.5. The results showed that there is similarity between some acetowhite temporal patterns of normal and abnormal tissues. Therefore we conclude that it is not sufficient to only consider the temporal dynamic of the acetowhitening reaction to establish a diagnosis by an automatic method. Information from cytologic, colposcopic and histopathologic disciplines should be integrated as well.

Improving Motor Imagery Classification With a New BCI Design Using Neuro-Fuzzy S-dFasArt

This paper presents an algorithm based on neural networks and fuzzy theory (S-dFasArt) to classify spontaneous mental activities from electroencephalogram (EEG) signals, in order to operate a noninvasive brain-computer interface. The focus is placed on the three-class problem, left-hand movement imagination, right movement imagination and word generation. The algorithm allows a supervised classification of temporal patterns improving the classification rates of the BCI Competition III (Data Set V: multiclass problem, continuous EEG). Using the precomputed data supplied for the competition and following the rules established there, a new method based on S-dFasArt, along with rule prune and voting strategy is proposed. The results have been compared with other published methods improving their success rates.

English

There are two paradigms for modelling the varying length temporal data namely, modelling the sequences of feature vectors as in the hidden Markov model-based approaches for speech recognition and modelling the sets of feature vectors as in the Gaussian mixture model (GMM)-based approaches for speech emotion recognition. In this paper, the methods using discrete hidden Markov models (DHMMs) in the kernel feature space and string kernel-based SVM classifier for classification of discretised representation of sequence of feature vectors obtained by clustering and vector quantisation in the kernel feature space are presented. The authors then present continuous density hidden Markov models (CDHMMs) in the explicit kernel feature space that use the continuous valued representation of features extracted from the temporal data. The methods for temporal pattern classification by mapping a varying length sequential pattern to a fixed-length sequential pattern and then using an SVM-based classifier for classification are also presented. The task of recognition of spoken letters in E-set, it is possible to build models that use a discretised representation and string kernel SVM based classification and obtain a classification performance better than that of models using the continuous valued representation is demonstrated. For modelling sets of vectors-based representation of temporal data, two approaches in a hybrid framework namely, the score vector-based approach and the segment modelling based approach are presented. In both approaches, a generative model-based method is used to obtain a fixed length pattern representation for a varying length temporal data and then a discriminative model is used for classification. These two approaches are studied for speech emotion recognition task. The segment modelling based approach gives a better performance than the score vector-based approach and the GMM-based classifiers for speech emotion recognition.Defence Science Journal, 2010, 60(4), pp.348-363, DOI:http://dx.doi.org/10.14429/dsj.60.492

Spiking Neural Networks for Cortical Neuronal Spike Train Decoding

Recent investigation of cortical coding and computation indicates that temporal coding is probably a more biologically plausible scheme used by neurons than the rate coding used commonly in most published work. We propose and demonstrate in this letter that spiking neural networks (SNN), consisting of spiking neurons that propagate information by the timing of spikes, are a better alternative to the coding scheme based on spike frequency (histogram) alone. The SNN model analyzes cortical neural spike trains directly without losing temporal information for generating more reliable motor command for cortically controlled prosthetics. In this letter, we compared the temporal pattern classification result from the SNN approach with results generated from firing-rate-based approaches: conventional artificial neural networks, support vector machines, and linear regression. The results show that the SNN algorithm can achieve higher classification accuracy and identify the spiking activity related to movement control earlier than the other methods. Both are desirable characteristics for fast neural information processing and reliable control command pattern recognition for neuroprosthetic applications.

Temporal pattern recognition using spiking neural networks for cortical neuronal spike train decoding

Most experimental and decoding algorithm studies of brain neural signals assume that neurons transmit information as a rate coding, but recent studies on the fast cortical computations indicate that temporal coding is probably a more biologically plausible scheme used by neurons. We introduce spiking neural networks (SNN) which consist of spiking neurons propagate information by the timing of spikes to analyze the cortical neural spike trains directly without temporal information lost. The SNN based temporal pattern classification is compared with the conventional artificial neural networks (ANN) based firing rate analysis. The results show that the SNN algorithm can achieve higher accuracy, which demonstrates that temporal coding is a viable code for fast neural information processing and the SNN approach is suitable for recognizing the temporal pattern in the cortical neural signals.

Quantitative Analysis of Dynamic Contrast-Enhanced MRI Datasets of the Metacarpophalangeal Joints

In this article, we propose an alternative approach to voxel-by-voxel analysis, which overcomes problems associated with heuristic methods currently used for dynamic contrast-enhanced MRI (DCE-MRI) data assessment. We aim to allow fully automated extraction of various heuristic parameters via robust preprocessing methods and a new technique for classification of temporal patterns of contrast agent uptake, making full use of all available dynamic frames of the datasets. We also demonstrate that application of efficient preprocessing methods permits more accurate analysis of the dynamic data. Ten DCE-MRI datasets enhanced by gadolinium diethylene triamine pentacetic acid were acquired from patients with rheumatoid arthritis using a 1.5-T MRI scanner. Fully automated voxel-by-voxel analysis of DCE-MRI signal intensity curves from 60 temporal slices was performed using a new method. Qualitative evaluation of the degree of inflammation was done via constructing parametric maps and quantitative by computing various heuristics such as maximum rate of enhancement, initial rate of enhancement, and time of onset of enhancement. Quantitative and qualitative evaluation obtained for 10 DCE-MRI datasets is presented. We demonstrate that preprocessing techniques compensate for patient movement, contribute to data fidelity and therefore permit more robust estimation of various heuristics such as maximum rate of enhancement, initial rate of enhancement, and time of onset of enhancement. Automatically generated parametric maps of these heuristics show favorable characteristics, permitting easier differentiation of structures of interest. These results are free from the subjective input and therefore easily reproducible. Furthermore, the proposed classification scheme provides information on the pattern of contrast uptake previously unavailable. Our preliminary results demonstrate the potential of the proposed method for providing objective quantitative and qualitative assessment of DCE-MRI in the metacarpophalangeal joints. Further evaluation within a clinical setting is needed to examine the method's diagnostic utility.

Temporal pattern identification using spike-timing dependent plasticity

This paper addresses the question of the functional role of the dual application of positive and negative Hebbian time dependent plasticity rules, in the particular framework of reinforcement learning tasks. Our simulations take place in a recurrent network of spiking neurons with inhomogeneous synaptic weights. A spike-timing dependent plasticity (STDP) rule is combined with its “opposite”, the “anti-STDP”. A local regulation mechanism moreover maintains the postsynaptic neuron in the vicinity of a reference frequency, which forces the global dynamics to be maintained in a softly disordered regime. This approach is tested on a simple discrimination task which requires short-term memory: temporal pattern classification. We show that such temporal patterns can be categorised, and present tracks for future improvements.

Neural network systems for multi-dimensional temporal pattern classification

Classification of multivariate, multi-class, temporal patterns is an important yet challenging problem that arises during state identification in agile processes. In this paper, we propose two new neural network structures that overcome the curse of dimensionality and the complexities in temporal pattern recognition. The one-variable-one-network (OVON) architecture decomposes the problem in the variable dimension—univariate temporal patterns, called sub-states, are first identified in each variable; subsequently, the process state is inferred using the variable sub-states. The one-class-one-network (OCON) architecture uses a problem decomposition in the class dimension—the presence of specific temporal patterns in a high-dimensional space is first established using a bank of two-class classifiers. The process state is then identified by aggregating the results of the different classifiers. Both the architectures use a set of neural networks—in OVON there is one network for each variable, while in OCON, one network is used for each pattern class to be identified. In comparison to traditional monolithic neural networks, both the proposed architectures improve classification accuracy and minimize the training complexity. In addition, OVON is robust to sensor failures and OCON is well suited for addition of new pattern classes. The structures and training methodologies of the two architectures are presented and their performance compared against traditional neural networks using patterns arising during transitions in a simulated fluidized catalytic cracking unit.

Classification of temporal patterns in dynamic biological networks.

A general method is presented to classify temporal patterns generated by rhythmic biological networks when synaptic connections and cellular properties are known. The method is discrete in nature and relies on algebraic properties of state transitions and graph theory. Elements of the set of rhythms generated by a network are compared using a metric that quantifies the functional differences among them. The rhythms are then classified according to their location in a metric space. Examples are given, and biological implications are discussed.

A signal processing framework based on dynamic neural networks with application to problems in adaptation, filtering, and classification

We present a coherent neural net based framework for solving various signal processing problems. It relies on the assertion that time-lagged recurrent networks possess the necessary representational capabilities to act as universal approximators of nonlinear dynamical systems. This applies to system identification, time-series prediction, nonlinear filtering, adaptive filtering, and temporal pattern classification. We address the development of models of nonlinear dynamical systems, in the form of time-lagged recurrent neural nets, which can be used without further training. We employ a weight update procedure based on the extended Kalman filter (EKF). Against the tendency for a net to forget earlier learning as it processes new examples, we develop a technique called multistream training. We demonstrate our framework by applying it to 4 problems. First, we show that a single time-lagged recurrent net can be trained to produce excellent one-time-step predictions for two different time series and also to be robust to severe errors in the input sequence. Second, we model stably a complex system containing significant process noise. The remaining two problems are drawn from real-world automotive applications. One involves input-output modeling of the dynamic behavior of a catalyst-sensor system which is exposed to an operating engine's exhaust stream, the other the real-time and continuous detection of engine misfire.

Cortical codes for sound location

The responses of auditory cortical neurons carry information both in the number of spikes within a burst and in the timing of those spikes. The capacity of single neurons to encode sound locations was studied in areas A1 and AES of α-chloralose-anesthetized cats. The spike counts of about half of studied neurons were modulated by more than 50% as a noise source was varied in azimuth. Nevertheless, the ‘‘best areas’’ of most units, using a 50%-of-maximum-response criterion, were larger than 180°. Temporal spike patterns of single neurons could be classified according to sound source azimuth, using an artificial neuron network. The classification of temporal patterns consistently surpassed a maximum-likelihood classification of spike counts. The spike patterns of single neurons could successfully encode locations throughout 360° of azimuth. The localization performance of any single neuron was substantially worse than that of an awake, behaving cat. When responses were averaged across multiple trials, however, performance improved monotonically with increases in the number of trials, suggesting that the information carried out by the temporal spike patterns of multiple neurons could account for a cat’s localization behavior.