Learning Epochs Research Articles

A Supervised Learning Algorithm for Spiking Neurons Using Spike Train Kernel Based on a Unit of Pair-Spike

In recent years, neuroscientists have discovered that the neural information is encoded by spike trains with precise times. Supervised learning algorithm based on the precise times for spiking neurons becomes an important research field. Although many existing algorithms have the excellent learning ability, most of their mechanisms still have some complex computations and certain limitations. Moreover, the discontinuity of spiking process also makes it very difficult to build an efficient algorithm. This paper proposes a supervised learning algorithm for spiking neurons using the kernel function of spike trains based on a unit of pair-spike. Firstly, we comprehensively divide the intervals of spike trains. Then, we construct an optimal selection and computation method of spikes based on the unit of pair-spike. This method avoids some wrong computations and reduces the computational cost by using each effective input spike only once in every epoch. Finally, we use the kernel function defined by an inner product operator to solve the computing problem of discontinue spike process and multiple output spikes. The proposed algorithm is successfully applied to many learning tasks of spike trains, where the effect of our optimal selection and computation method is verified and the influence of learning factors such as learning kernel, learning rate, and learning epoch is analyzed. Moreover, compared with other algorithms, all experimental results show that our proposed algorithm has the higher learning accuracy and good learning efficiency.

Research and implementation of NN based on TensorFlow

TensorFlow is Google’s open-source machine learning and deep learning framework, which is convenient and flexible to build the current mainstream deep learning model. Convolutional neural network is a classical model of deep learning, the advantage lies in its powerful feature extraction capabilities of convolutional blocks. A neural network in the simplest case is a mathematical model consisting of several layers of elements that perform parallel calculations. Initially, such an architecture was created by analogy with the small computing elements of the human brain — neurons. The minimal computing elements of an artificial neural network are also called neurons. Neural networks typically consist of three or more layers: an input layer, a hidden layer (or layers), and an output layer. An important feature of the neural network is its ability to learn by example, this is called learning with a teacher. The neural network is trained on a large number of examples consisting of input-output pairs (corresponding to each other input and output). In object recognition problems, such a pair will be the input image and the corresponding label — the name of the object. Neural network learning is an iterative process that reduces the deviation of the network output from a given «teacher response» — a label that corresponds to a given image. This process consists of steps called epochs of learning (they are usually calculated in thousands), each of which is the adjustment of the «weights» of the neural network — the parameters of the hidden layers of the network. Upon completion of the learning process, the quality of the neural network is usually good enough to perform the task for which it was trained, although the optimal set of parameters that perfectly recognizes all the images, it is often impossible to choose. Based on the TensorFlow platform, a convolutional neural network model with two-convolution-layers was built. The model was trained and tested with the MNIST data set. The test accuracy rate could reach 99,15%, and compared with the rate of 98,69% with only one-convolution-layer model, which shows that the two-convolution-layers convolutional neural network model has a better ability of feature extraction and classification decision-making.

TensorFlow Quantum: Impacts of Quantum State Preparation on Quantum Machine Learning Performance

Learning methodologies on quantum devices have shown that there are advantages in utilizing quantum properties. A requirement for using quantum computing in machine learning techniques is the data representation as quantum states. In Quantum Machine Learning, quantum state preparation is paramount to attain a functional pipeline in a model. One state preparation method, amplitude encoding, allows a dataset to be mapped or encoded more robustly and enhances the learning of quantum models. Albeit more densely represented, a dataset which has been prepared by amplitude encoding provides a more learnable input to a model. The two main advantages from using amplitude encoding are an increase in classification accuracy and reduced variability of learning epoch to epoch. In this paper, we compare the basic implementations of TensorFlow Quantum’s Quantum Convolutional Neural Network and a hybrid quantum-classical network using angle encoding, with a third network of our design that utilizes amplitude encoding for enriched state preparation. Our results show there is a direct benefit in performing amplitude encoding before training a TensorFlow Quantum hybrid quantum-classical model. In the best case scenario, amplitude encoding made classifying the samples 8.9% more accurate.

Stable polyp-scene classification via subsampling and residual learning from an imbalanced large dataset.

This Letter presents a stable polyp-scene classification method with low false positive (FP) detection. Precise automated polyp detection during colonoscopies is essential for preventing colon-cancer deaths. There is, therefore, a demand for a computer-assisted diagnosis (CAD) system for colonoscopies to assist colonoscopists. A high-performance CAD system with spatiotemporal feature extraction via a three-dimensional convolutional neural network (3D CNN) with a limited dataset achieved about 80% detection accuracy in actual colonoscopic videos. Consequently, further improvement of a 3D CNN with larger training data is feasible. However, the ratio between polyp and non-polyp scenes is quite imbalanced in a large colonoscopic video dataset. This imbalance leads to unstable polyp detection. To circumvent this, the authors propose an efficient and balanced learning technique for deep residual learning. The authors’ method randomly selects a subset of non-polyp scenes whose number is the same number of still images of polyp scenes at the beginning of each epoch of learning. Furthermore, they introduce post-processing for stable polyp-scene classification. This post-processing reduces the FPs that occur in the practical application of polyp-scene classification. They evaluate several residual networks with a large polyp-detection dataset consisting of 1027 colonoscopic videos. In the scene-level evaluation, their proposed method achieves stable polyp-scene classification with 0.86 sensitivity and 0.97 specificity.

NIMG-11. VISUALIZATION OF JUDGMENT BASIS OF CNN TO GRADING GLIOMA

Abstract BACKGROUND convolutional neural network (CNN) model using MRI imaging are likely to be effective for grading glioma. However, interpretation of the judgment basis of CNN is difficult. The purpose of this study is twofold. One is to create a high accuracy machine learning model that grading of glioma (Glioma grade II, between III and IV). The other is to visualize the judgment basis of the model. METHODS We targeted cases that were imaged at our Hospital during the period from August 2014 to January 2018. Five types of MRI are used. The five types are two types of DWI (b1000DWI and b2000DWI, respectively, with a value of 1000 and 2000), apparent diffusion coefficient (ADC), fractional anisotropy (FA), and mean kurtosis (MK). The images were input to CNN wtihout ROI. Seven types of CNN were prepared, and 100 epochs of learning were performed. RESULTS 55 cases were included. There were 14 grade II, 12 grade III, and 29 grade IV). Of these, 44 cases up to July 2017 in chronological order were used as a dataset for learning, and 11 cases from August 2017 to January 2018 were taken as independent test dataset. The best results were obtained using MK as input, with a percentage of correct cases of test data of 0.82. Subsequently, we applied Grad-CAM to the model. Grad-CAM is visualized technique that shows where CNN focused on. The model focused on the tumor part of the image. In addition, in the cases of glioma and meningioma coexisting, grading was performed focusing on the glioma area not on meningioma area. CONCLUSION When grading glioma, it was considered that CNN decide glioma’s grade focusing on tumor part of the image, as well as humans.

Near Optimal Coded Data Shuffling for Distributed Learning

Data shuffling between distributed cluster of nodes is one of the critical steps in implementing large-scale learning algorithms. Randomly shuffling the data-set among a cluster of workers allows different nodes to obtain fresh data assignments at each learning epoch. This process has been shown to provide improvements in the learning process (via testing and training error). However, the statistical benefits of distributed data shuffling come at the cost of extra communication overhead from the master node to worker nodes, and can act as one of the major bottlenecks in the overall time for computation. There has been significant recent interest in devising approaches to minimize this communication overhead. One approach is to provision for extra storage at the computing nodes. The other emerging approach is to leverage coded communication to minimize the overall communication overhead. The focus of this work is to understand the fundamental tradeoff between the amount of storage and the communication overhead for distributed data shuffling. In this paper, we first present an information theoretic formulation for the data shuffling problem, accounting for the underlying problem parameters (number of workers, $K$ , number of data points, $N$ , and available storage, and $S$ per node). We then present an information theoretic lower bound on the communication overhead for data shuffling as a function of these parameters. We next present a novel coded communication scheme and show that the resulting communication overhead of the proposed scheme is within a multiplicative factor of at most $\frac {K}{K-1}$ from the lower bound (which is upper bounded by 2 for $K\geq 2$ ). Furthermore, we present new results towards closing this gap through a novel coded communication scheme, which we call the aligned coded shuffling. This scheme is inspired by the ideas of coded shuffling and interference alignment. In particular, we show that the aligned scheme achieves the optimal storage vs communication trade-off for $K , and further reduces the maximum multiplicative gap down to $\frac {K-\frac {1}{3}}{K-1}$ , for $K\geq 5$ .

A Delay Learning Algorithm Based on Spike Train Kernels for Spiking Neurons.

Neuroscience research confirms that the synaptic delays are not constant, but can be modulated. This paper proposes a supervised delay learning algorithm for spiking neurons with temporal encoding, in which both the weight and delay of a synaptic connection can be adjusted to enhance the learning performance. The proposed algorithm firstly defines spike train kernels to transform discrete spike trains during the learning phase into continuous analog signals so that common mathematical operations can be performed on them, and then deduces the supervised learning rules of synaptic weights and delays by gradient descent method. The proposed algorithm is successfully applied to various spike train learning tasks, and the effects of parameters of synaptic delays are analyzed in detail. Experimental results show that the network with dynamic delays achieves higher learning accuracy and less learning epochs than the network with static delays. The delay learning algorithm is further validated on a practical example of an image classification problem. The results again show that it can achieve a good classification performance with a proper receptive field. Therefore, the synaptic delay learning is significant for practical applications and theoretical researches of spiking neural networks.

Dynamic Modification of Activation Function using the Backpropagation Algorithm in the Artificial Neural Networks

The paper proposes the dynamic modification of the activation function in a learning technique, more exactly backpropagation algorithm. The modification consists in changing slope of sigmoid function for activation function according to increase or decrease the error in an epoch of learning. The study was done using the Waikato Environment for Knowledge Analysis (WEKA) platform to complete adding this feature in Multilayer Perceptron class. This study aims the dynamic modification of activation function has changed to relative gradient error, also neural networks with hidden layers have not used for it.

Online Sequential Extreme Learning Machine With Dynamic Forgetting Factor

Online sequential extreme learning machine (OS-ELM) and its variants provide a promising way to solve data stream problems, but most of them do not take the timeliness of the problems into account, which may degrade the performance of the model. The main reason is that these algorithms are unable to adapt to the latest data accordingly when the distribution of the data stream changes. To mitigate this limitation, the forgetting factor is introduced into the relevant models, which is used to balance the relative importance of past data and new data when necessary. However, there is no efficient way to set the forgetting factor properly so far. In this paper, we have developed a novel updating strategy for setting the forgetting factor and proposed a dynamic forgetting factor based OS-ELM algorithm (DOS-ELM). In the sequential learning phase of DOS-ELM, the forgetting factor can be adjusted dynamically according to the change degree of the model accuracy in each learning epoch. This updating process does not require setting any parameters artificially and thus greatly improves the flexibility of the model. The experimental results on ten classification problems, five regression problems, one time-series problem show that DOS-ELM can deal well with both stationary and non-stationary data stream problems. In addition, we have extended DOS-ELM to an online deep model named ML-DOS-ELM, which can handle more complex tasks such as the face recognition problem and the handwritten digit recognition problem. Our experimental evaluations show that both DOS-ELM and ML-DOS-ELM can achieve higher prediction accuracy compared to the other similar algorithms.

THE USE OF A NEURAL NETWORK FOR IDENTIFICATION OF BURNOUT ZONES IN DIAGNOSTICS OF THE LINING OF CRITICAL LINED EQUIPMENT

A method of automated diagnostics of critical lined equipment is improved using a neural network to recognize the thermograms and classify the burnout zones. The proposed method provides an increase in the reliability and promptness in determination of the lining burnout zones and their qualitative assessment. The information signs used for analysis and recognition of the lining burnout zones on a thermogram image are considered. The results of using neural networks with different configurations to minimize the error of classifying the levels of the burnout and determining the optimal number of the learning epochs are presented. The developed automated system for technical diagnostics of critical lined equipment, including the software for analysis and recognition of the thermograms, was evaluated and implemented at the enterprises of metallurgical production. Comparative analysis of the results obtained using the developed automated system and traditional system of diagnostics demonstrated the advantages of the developed method.

Surveillance Planning With Bézier Curves

This letter concerns surveillance planning for an unmanned aerial vehicle (UAV) that is requested to periodically take snapshots of areas of interest by visiting a given set of waypoint locations in the shortest time possible. The studied problem can be considered as a variant of the combinatorial traveling salesman problem in which trajectories between the waypoints respect the kinematic constraints of the UAV. Contrary to the existing formulation for curvature-constrained vehicles known as the Dubins traveling salesman problem, the herein addressed problem is motivated by planning for multirotor UAVs which are not limited by the minimal required forward velocity and minimal turning radius as the Dubins vehicle, but rather by the maximal speed and acceleration. Moreover, the waypoints to be visited can be at different altitudes, and the addressed problem is to find a fast and smooth trajectory in three-dimensional (3-D) space from which all the areas of interest can be captured. The proposed solution is based on unsupervised learning in which the requested 3-D smooth trajectory is determined as a sequence of Bezier curves in a finite number of learning epochs. The reported results support feasibility of the proposed solution which has also been experimentally verified with a real UAV.

Prediction of Availability Indicator of Water Pipes Using Artificial Intelligence

The paper presents the results of artificial neural networks application to the availability indicator prediction. The forecasted results indicate that artificial networks may be used to model the reliability level of the water supply systems. The network was trained using 147 and 173 operational data from one Polish medium-sized city (distribution pipes and house connections, respectively). 50% of all data was chosen for learning, 25% for testing and 25% for validation. In prognosis phase, the best created network used 100% of 114 and 133 values for testing. Following functions were used to activate neurons in hidden and output layers: linear, logistic, hyperbolic tangent, exponential. The learning of the artificial network was performed using following input parameters: material, total length, diameter. In the optimal models hyperbolic tangent was chosen to activate the hidden and output neurons in modeling the availability indicator of house connections during 68 epochs of training. Hidden and output neurons were activated (20 epochs of learning) respectively by hyperbolic tangent and linear function during the prediction of availability indicator of distribution pipes. The maximum relative errors in learning and prognosis step were equal to 0.10% and 1.20% as well as 0.27% and 1.15% for distribution pipes and house connections, respectively.

Adaptive Scaling of Cluster Boundaries for Large-Scale Social Media Data Clustering.

The large scale and complex nature of social media data raises the need to scale clustering techniques to big data and make them capable of automatically identifying data clusters with few empirical settings. In this paper, we present our investigation and three algorithms based on the fuzzy adaptive resonance theory (Fuzzy ART) that have linear computational complexity, use a single parameter, i.e., the vigilance parameter to identify data clusters, and are robust to modest parameter settings. The contribution of this paper lies in two aspects. First, we theoretically demonstrate how complement coding, commonly known as a normalization method, changes the clustering mechanism of Fuzzy ART, and discover the vigilance region (VR) that essentially determines how a cluster in the Fuzzy ART system recognizes similar patterns in the feature space. The VR gives an intrinsic interpretation of the clustering mechanism and limitations of Fuzzy ART. Second, we introduce the idea of allowing different clusters in the Fuzzy ART system to have different vigilance levels in order to meet the diverse nature of the pattern distribution of social media data. To this end, we propose three vigilance adaptation methods, namely, the activation maximization (AM) rule, the confliction minimization (CM) rule, and the hybrid integration (HI) rule. With an initial vigilance value, the resulting clustering algorithms, namely, the AM-ART, CM-ART, and HI-ART, can automatically adapt the vigilance values of all clusters during the learning epochs in order to produce better cluster boundaries. Experiments on four social media data sets show that AM-ART, CM-ART, and HI-ART are more robust than Fuzzy ART to the initial vigilance value, and they usually achieve better or comparable performance and much faster speed than the state-of-the-art clustering algorithms that also do not require a predefined number of clusters.

Extending the tempotron with hierarchical dendrites allows faster learning

Certain functional classes of neurons seem to be able to differentiate between input patterns with high temporal precision. As input patterns to neurons can consist of up to thousands of inputs, the ability to identify target patterns amongst statistically similar background patterns is impressive and has been suggested to occur via modification to synaptic weights. Yet how does learning that occurs locally at the level of synapses converge, without global coordination? While this question is important for understanding how synaptic learning gives rise to dendritic computation, it is impractical to test experimentally. Insight from abstract neuronal models, such as the multilayer perceptron networks [1], provide a potential glimpse of the difficulty of how to ensure that global convergence during learning when weight changes are local. In this work, we report that by extending the tempotron model [2], we are able to demonstrate that learning can, indeed, learn locally but also converge globally. By arranging dendritic units in a hierarchical manner which feeds into a master dendrite and soma, learning occurs over two timescales: locally on each dendritic branch, using a simple incremental plasticity rule; and at a slower timescale on the main branch, where information is integrated across branches. We observe that the inclusion of dendrites reduces the learning time required by allowing dendrites to subsample the entire input space. In comparison to one single dendrite receiving n inputs, the inclusion of m dendrites means that each dendrite is now subsampling n/m inputs, which not results in faster learning epochs before convergence but also improves the overall robustness against noise. Thus, the move from a single tempotron to a set of hierarchically configured tempotrons, representing dendrites, imbues the unit with recognition of pattern fragments (with a pattern capacity > m!), faster convergence during learning and increased noise tolerance (both of which scale with m). The inclusion of dendrites also allows for them to signal in sequences with varying relative temporal offsets, granting the neuron the opportunity to differentiate between multiple positive patterns to identify not only which pattern was observed but also when. Furthermore, we have also demonstrated that tempotrons can be extended to work for non-episodic patterns i.e. ongoing and without reset, and can also perform well when number of distractor patterns greatly outnumber positive patterns. Conceptually, our model reconciles the tempotron learning rule with work on dendritic computation which argues for dendrites as computation units [3]. It also provides a potential explanation to phenomena observed experimentally, such as neurons in visual cortex whose dendrites had other preferred orientations that were different to those of the soma [4].

Deficits in hippocampal-dependent transfer generalization learning accompany synaptic dysfunction in a mouse model of amyloidosis.

Elevated β-amyloid and impaired synaptic function in hippocampus are among the earliest manifestations of Alzheimer's disease (AD). Most cognitive assessments employed in both humans and animal models, however, are insensitive to this early disease pathology. One critical aspect of hippocampal function is its role in episodic memory, which involves the binding of temporally coincident sensory information (e.g., sights, smells, and sounds) to create a representation of a specific learning epoch. Flexible associations can be formed among these distinct sensory stimuli that enable the "transfer" of new learning across a wide variety of contexts. The current studies employed a mouse analog of an associative "transfer learning" task that has previously been used to identify risk for prodromal AD in humans. The rodent version of the task assesses the transfer of learning about stimulus features relevant to a food reward across a series of compound discrimination problems. The relevant feature that predicts the food reward is unchanged across problems, but an irrelevant feature (i.e., the context) is altered. Experiment 1 demonstrated that C57BL6/J mice with bilateral ibotenic acid lesions of hippocampus were able to discriminate between two stimuli on par with control mice; however, lesioned mice were unable to transfer or apply this learning to new problem configurations. Experiment 2 used the APPswe PS1 mouse model of amyloidosis to show that robust impairments in transfer learning are evident in mice with subtle β-amyloid-induced synaptic deficits in the hippocampus. Finally, Experiment 3 confirmed that the same transfer learning impairments observed in APPswePS1 mice were also evident in the Tg-SwDI mouse, a second model of amyloidosis. Together, these data show that the ability to generalize learned associations to new contexts is disrupted even in the presence of subtle hippocampal dysfunction and suggest that, across species, this aspect of hippocampal-dependent learning may be useful for early identification of AD-like pathology.

Data Clustering using Memristor Networks.

Memristors have emerged as a promising candidate for critical applications such as non-volatile memory as well as non-Von Neumann computing architectures based on neuromorphic and machine learning systems. In this study, we demonstrate that memristors can be used to perform principal component analysis (PCA), an important technique for machine learning and data feature learning. The conductance changes of memristors in response to voltage pulses are studied and modeled with an internal state variable to trace the analog behavior of the device. Unsupervised, online learning is achieved in a memristor crossbar using Sanger’s learning rule, a derivative of Hebb’s rule, to obtain the principal components. The details of weights evolution during training is investigated over learning epochs as a function of training parameters. The effects of device non-uniformity on the PCA network performance are further analyzed. We show that the memristor-based PCA network is capable of linearly separating distinct classes from sensory data with high clarification success of 97.6% even in the presence of large device variations.

Neural Network Structure Analysis Based on Hierarchical Force-Directed Graph Drawing for Multi-Task Learning

A hierarchical force-directed graph drawing is proposed for the analysis of a neural network structure that expresses the relationship between multitask and processes in neural networks represented as neuron clusters. The process revealed by our proposal indicates the neurons that are related to each task and the number of neurons or learning epochs that are sufficient. Our proposal is evaluated by visualizing neural networks learned on the Mixed National Institute of Standards and Technology (MNIST) database of handwritten digits, and the results show that inactive neurons, namely those that do not have a close relationship with any tasks, are located on the periphery part of the visualized network, and that cutting half of the training data on one specific task (out of ten) causes a 15% increase in the variance of neurons in clusters that react to the specific task compared to the reaction to all tasks. The proposal aims to be developed in order to support the design process of neural networks that consider multitasking of different categories, for example, one neural network for both the vision and motion system of a robot.

Application of Artificial Neural Network Model to Human Body Vibrations in Large Haul Trucks

The working conditions in oil sand mines in Northern Alberta, Canada, are greatly affected by the climate and geology. The ground is very hard and competent in winter but very soft during summer [1]. This changing behaviour of the ground has a great impact on the truck’s frame and the health of the operator because he is exposed to large Whole Body Vibrations (WBV). When the human body is exposed to large WBV for prolonged periods, it begins to have chronic back problems resulting in diseases of the lumbar spine, disc degeneration and other pathological effects to the spine and skeletal structure [2]. Many of these health effects are irreversible and people suffering from WBV disorders can experience pains for the rest of their lives. Studies carried out to correlate WBV to health problems have resulted in the setting of standards with the help of various government agencies, physiologists and industry experts. The main WBV standards that the mining industry has to observe are those of International Standards Organization (ISO) and the British Standards Institute [2]. ISO 2631-1 uses a three dimensional coordinate system where the axes are orthogonal to each other as shown in Figure 1 [3]. The standards require that the measuring device (accelerometer) should be placed at a point where vibrations are entering the body. In this study the data was collected for the seat accelerations in X, Y and Z directions by installing the accelerometer on the seat pan of a CAT 797 haul truck. An Artificial Neural Network (ANN) model was developed to predict the seat vibrations in very large capacity haul trucks in the X and Y directions. The study was done to find if the vibrations in the truck and their effects on the operator can be correlated to the truck’s operational parameters like speed, payload and strut pressures. There is a need for an onboard monitoring system on large haul trucks to alert the driver when he exceeds specified speed limits at certain locations along the haul road which are unsafe for him. Caterpillar 797 trucks use Vital Information Management System (VIMS) to track the truck’s behaviour when driven along a haul road. It includes onboard truck measuring equipment and off-board VIMS software which enables data logging and downloading into a computer. A record of real time parameters gives a detailed view of what happens to the truck body while performing the various functions [4]. The disadvantage of using a piezo-resistive accelerometer to collect the data collection is that it has a limited high frequency response [5]. Thus, it is necessary that the data be collected at higher frequencies to obtain a better model that can give more accurate predictions. In this work the accelerometer was replaced with an ANN model. The ANN model is developed for the dump truck such that it models the seat vibrations in the truck in response to the truck’s speed, payload, changing ground stiffness and profile. These truck parameters can be estimated from the strut pressure responses measured in terms of machine rack, machine roll and machine pitch. The machine’s roll, rack and pitch give an idea of the truck’s response to the changing ground conditions and operator’s driving skills. These values are calculated using the changing pressures of the four struts on which the whole truck frame is resting. These parameters were used as input variables during the development of the ANN model. This study focused only on the seat accelerations in the X-direction and Y-direction as output variables, as these are more important than the accelerations in the Z-direction. The ANN model is very efficient for non-linear processes and tends to improve its performance as it learns more and more about the system. Thus when the ANN model is used, very accurate acceleration values can be predicted. This is important for the health of the operator and also for the lifespan of the truck as the high amplitude vibrations are detrimental to truck’s structural components. Using the NeuroShell® 2 software, a multilayer perceptron ANN was trained with the back propagation algorithm. The model’s performance was optimized following the systematic approach developed by [6,7]. This is done by looking for the simplest network architecture that can converge. During the development of the ANN model for Seat X acceleration, the number of hidden layer neurons was increased from 10 to 70 while keeping the number of learning epochs constant at 50. The number of neurons which gave the best value for R 2 (Validation)

Prediction of radical scavenging activities of anthocyanins applying adaptive neuro-fuzzy inference system (ANFIS) with quantum chemical descriptors.

Radical scavenging activity of anthocyanins is well known, but only a few studies have been conducted by quantum chemical approach. The adaptive neuro-fuzzy inference system (ANFIS) is an effective technique for solving problems with uncertainty. The purpose of this study was to construct and evaluate quantitative structure-activity relationship (QSAR) models for predicting radical scavenging activities of anthocyanins with good prediction efficiency. ANFIS-applied QSAR models were developed by using quantum chemical descriptors of anthocyanins calculated by semi-empirical PM6 and PM7 methods. Electron affinity (A) and electronegativity (χ) of flavylium cation, and ionization potential (I) of quinoidal base were significantly correlated with radical scavenging activities of anthocyanins. These descriptors were used as independent variables for QSAR models. ANFIS models with two triangular-shaped input fuzzy functions for each independent variable were constructed and optimized by 100 learning epochs. The constructed models using descriptors calculated by both PM6 and PM7 had good prediction efficiency with Q-square of 0.82 and 0.86, respectively.

CHAOTIC FEATURE EXTRACTION AND NEURO-FUZZY CLASSIFIER FOR ECG SIGNAL CHARACTERIZATION

In this paper, a neuro-fuzzy network is employed to classify the ECG beats based on the extracted chaotic features. Six groups of ECG beats (MIT-BIH Normal Sinus rhythm, BIDMC congestive heart failure, CU ventricular tachyarrhythmia, MIT-BIH atrial fibrillation, MIT-BIH Malignant Ventricular Arrhythmia and MIT-BIH supraventricular arrhythmia) are characterized by the six chaotic parameters including the largest Lyapunov exponent and average of the Lyapunov spectrum (related to the chaoticity of the signal), time lag and embedding dimension (related to the phase space reconstruction) and correlation dimension and approximate entropy of the signal (related to the complexity of the signal). Finally, six structures of the neuro-fuzzy network (in terms of the type of fuzzy set, the number of fuzzy sets per variable and the number of learning epochs) were employed to perform the ECG beats classification based on all extracted features for two lengths of the signals. It was found that all respective chaotic features are discriminative and they improve the classification rate of ECG beats. Also, it is shown that a minimum length of the signal is needed for exhibitive feature extraction and for the higher lengths of the signal (in time) no significant improvement is achieved in feature extraction and calculations. The criteria for the classification task are considered as accuracy, specificity and sensitivity which all together comprehensively demonstrate the capability and performance of the classification. Some conclusions are drawn and they are discussed at the end of the paper.