Complex Learning Algorithms Research Articles

Efficient Vehicle Detection and Optimization in Multi-Graph Mode Considering Multi-Section Tracking Based on Geographic Similarity

Vehicle detection is an important part of modern intelligent transportation systems. At present, complex deep learning algorithms are often used for vehicle detection and tracking, but high-precision detection results are often obtained at the cost of time, and the existing research rarely considers optimization algorithms for vehicle information. Based on this, we propose an efficient method for vehicle detection in multi-graph mode and optimization method considering multi-section tracking based on geographic similarity. In this framework, we design a vehicle extraction method based on multi-graph mode and a vehicle detection technology based on traffic flow characteristics, which can cope with the challenge of vehicle detection under an unstable environment. Further, a multi-section tracking optimization technology based on geographic similarity at a high video frame rate is proposed, which can efficiently identify lane change behavior and match, track, and optimize vehicles. Experiments are carried out on several road sections, and the model performance and optimization effect are analyzed. The experimental results show that the vehicle detection and optimization algorithm proposed in this paper has the best effect and high detection accuracy and robustness. The average results of Recall, Precision, and F1 are 0.9715, 0.979, and 0.9752, respectively, all of which are above 0.97, showing certain competitiveness in the field of vehicle detection.

Classification and evaluation of the algorithms for vector bin packing

Heuristics for Vector Bin Packing (VBP) play an important role in modern distributed computing systems and other applications aimed at optimizing the usage of multidimensional resources. In this paper we perform a systematic classification of heuristics for VBP, with the focus on construction heuristics. We bring together existing VBP algorithms and their tuning parameters, and propose new algorithms and new tuning parameters. For a less studied class of multi-bin algorithms, we explore their properties analytically, considering monotonic and anomalous behavior and approximation guarantees. For empirical evaluation, all algorithms are implemented as the Vectorpack library and assessed through extensive experiments. Our findings may serve as the basis for the development of more complex, hybrid algorithms, hyperheuristics and machine learning algorithms. The Vectorpack library can also be adjusted for addressing enhanced VBP problems with additional features, which arise in applications, especially those typical for modern distributed computing systems.

Evaluating the Accuracy of a Linear Regression Model in Predicting the Dissolution of Tablets Based on Raman Maps

Investigation of the dissolution of tablets is an important area of pharmaceutical research. Such research aims to predict the dissolution process as accurately as possible without destroying the tablets. Several methods have been published that can estimate dissolution with approximate accuracy, but they are primarily complex learning algorithms that are time-consuming and require many samples to train. This article seeks to answer whether these complex models are necessary or whether a similar result can be achieved with the help of more straightforward methods. Therefore, during this work, a simpler linear regression model was created and analysed its effectiveness in estimating the dissolution curves. The investigation concluded that the results are not as accurate as in the case of more complex methods, but the model is more robust and can be used in the case of fewer samples. Thus, by further developing these and combining the methods, we can achieve better results in the future.

A Learning-Based Stable Servo Control Strategy Using Broad Learning System Applied for Microrobotic Control.

As the controller parameter adjustment process is simplified significantly by using learning algorithms, the studies about learning-based control attract a lot of interest in recent years. This article focuses on the intelligent servo control problem using learning from desired demonstrations. Compared with the previous studies about the learning-based servo control, a control policy using the broad learning system (BLS) is developed and first applied to a microrobotic system, since the advantages of the BLS, such as simple structure and no-requirement for retraining when new demos' data is provided. Then, the Lyapunov theory is skillfully combined with the complex learning algorithm to derive the controller parameters' constraints. Thus, the final control policy not only can obtain the movement skills of the desired demonstrations but also have the strong ability of generalization and error convergence. Finally, simulation and experimental examples verify the effectiveness of the proposed strategy using MATLAB and a microswimmer trajectory tracking system.

Student performance prediction with BPSO feature selection and CNN classifier

Educational Data Mining (EDM) is gaining great importance as a new interdisciplinary research field related to some other areas. It is directly related to data mining (DM), the latter being a fundamental part of knowledge discovery in databases (KDD). This data is growing more and more and contains hidden knowledge that could be very useful for users (both teachers and students). It is convenient to identify such knowledge in the form of models, patterns, or any other representation scheme that allows better exploitation of the system. Data mining is revealed as the tool to achieve such discovery, giving rise to EDM. In this complex context, different techniques and learning algorithms are usually used to obtain the best results. Recently educational systems are adopting artificial intelligent systems, especially in the educational context, specific areas for extracting relevant information, such as EDM, which integrates numerous techniques that support the capture, processing, and analysis of these sets of records. The main technique associated with EDM is Machine Learning, which has been used for decades in data processing in different contexts, but with the advent of Big Data, there was an intensification in the application of this technique to extract relevant information from a huge amount of data. This paper proposes the student performance prediction using CNN (Convolution Neural Network) and BPSO (Binary Particle Swarm Optimization) based feature selection method. In this study, classifiers are made for 2-class and 5-class predictions. The proposed system claims an outperforming accuracy of 96.6% with various previous research works as well as found that the majority of attributes related to school activities as compared to data on demographic and socioeconomic characteristics.

Adaptively Equalized Bandwidth Optimization Model using SCADA-DWWAN based Neural Network

Artificial training and learning algorithms, enhanced with semi-supervised or self-supervised feature extraction capacities, employ adaptive decision optimization models. These are often favored over complex deep learning algorithms for achieving better controllability and ease of observation, lower complexity in simulating, building or designing and virtual prototyping of automatic network resource management (ANRM) standards. An Adaptive Linear Neuron type Artificial Neural Network (ADALINE-ANN) which is based on multi-tapered machine learning approach has been simulated in a virtual Supervisory Control and Data Acquisition (SCADA) framework integrated with a Distributed Control System (SCADA/DCS-Net). The system has been virtually simulated considering an adaptively equalized learning and decision approach which utilizes Markov Trained-Steepest Gradient Descent (HMM-SGD) based machine learning model employing Kalman optimization. Affinity clustering is employed for spectrum sensing by extracting the Constellation Nyquist Bands from an M-Quadrature Amplitude Modulation (QAM) orthogonal signal undergoing AWGN and Rayleigh fading as well as co-channel interference (CCI), and ensemble analysis using Channel State-Space plots are used for optimal spectrum allocation in an Adaptive Orthogonal Frequency Division Multiple Access (Adaptive-OFDMA) layout. It has been done by implementing an adaptively equalized Automatic Repeat Request (ARQ) pipelining model which utilizes minimum least square error (MLSE) minimization model. The objective is to improve the bandwidth allocation and usage ensuring most minimum spectrum wastage or loss. Successive Interference Cancellation (SIC) has been implemented to minimize static buffer and interference loss. Thus, spectrum loss due to latency and jitter which occurs from bandwidth congestion is minimized by improving the network resource tracking and allocation. It results in improved and stable bandwidth equalization.

Development of a Novel Cavitation Monitoring System for Hydro Turbines Based on Machine Learning Algorithms

Operating hydro turbines in off-design conditions increases the risk of cavitation occurrence, which in turn, leads to numerous problems such as performance degradations, structural vibrations, and most importantly, mechanical damage due to erosion. It is therefore crucial to develop a monitoring system that detects the occurrence and severity of cavitation in real time. For this purpose, a cavitation detection methodology has been developed that is based on the analysis of acoustic emissions of a turbine with machine learning algorithms. In this method, a conventional microphone is used to record the airborne noise emitted from a turbine under different working conditions, and then, a supervised learning algorithm is trained to classify the recorded noise signals into cavitating and non-cavitating categories. The detection system was developed based on laboratory tests and was validated in Ernen hydropower plant located in Canton of Wallis in southeast of Switzerland. This power plant consists of two identical double-flux Francis turbines each having a maximum power of 16 MW and a net head of 270 mWC. The preliminary results obtained from a two-day experimental campaign in the Ernen powerplant are very promising in terms of cavitation detection with a classification accuracy of more than 90 %. The system could be implemented either for real-time monitoring of cavitation occurrence allowing the operators to avoid such a condition or as a post processing tool to evaluate the number of hours a turbine has worked under severe conditions. Work is still ongoing to deploy more complex learning algorithms for this task to minimize expert intervention and/or interpretation during the setup process.

Smart City Construction and Management by Digital Twins and BIM Big Data in COVID-19 Scenario

With the rapid development of information technology and the spread of Corona Virus Disease 2019 (COVID-19), the government and urban managers are looking for ways to use technology to make the city smarter and safer. Intelligent transportation can play a very important role in the joint prevention. This work expects to explore thebuilding information modeling (BIM) big data (BD)processing method ofdigital twins (DTs)of Smart City, thus speeding up the construction of Smart City and improve the accuracy of data processing. During construction, DTs build the same digital copy of the smart city. On this basis, BIM designs the building's keel and structure, optimizing various resources and configurations of the building. Regarding the fast data growth in smart cities, a complex data fusion and efficient learning algorithm, namely Multi-Graphics Processing Unit (GPU), is proposed to process the multi-dimensional and complex BD based on the compositive rough set model. The Bayesian network solves the multi-label classification. Each label is regarded as a Bayesian network node. Then, the structural learning approach is adopted to learn the label Bayesian network's structure from data. On the P53-old and the P53-new datasets, the running time of Multi-GPU decreases as the number of GPUs increases, approaching the ideal linear speedup ratio. With the continuous increase of K value, the deterministic information input into the tag BN will be reduced, thus reducing the classification accuracy. When K = 3, MLBN can provide the best data analysis performance. On genbase dataset, the accuracy of MLBN is 0.982 ± 0.013. Through experiments, the BIM BD processing algorithm based onBayesian Network Structural Learning (BNSL)helps decision-makers use complex data in smart cities efficiently.

An Input Weights Dependent Complex-Valued Learning Algorithm Based on Wirtinger Calculus

Complex-valued neural network is a kind of learning model which can deal with problems in complex domain. Fully complex extreme learning machine (CELM) is a much faster training algorithm than the complex backpropagation (CBP) scheme. However, it is at the cost of using more hidden nodes to obtain the comparable performance. An upper-layer-solution-aware algorithm has been proposed for training single-hidden layer feedforward neural networks, which performs much better than its counterparts, pseudo-inverse learning (PIL)/extreme learning machine and gradient decent-based backpropagation neural networks. Consequently, there exist two challenges that need to be dealt with: 1) How to combine the advantages of CBP and CELM to develop a novel complex learning algorithm? and 2) What is the convergent behavior of the presented algorithm? In this article, an input weights dependent complex-valued (IWDCV) learning algorithm based on Wirtinger calculus has been proposed, which effectively solves the nonanalytic problem of the common activation functions during training neural networks. In addition, the monotonicity of the error function and the deterministic convergence of the proposed model have been strictly proved, which theoretically guarantee the efficiency and effectiveness of the given model, IWDCV. Finally, for real and complex-valued problems, a variety of simulations have been done to demonstrate the comparable performance of the proposed algorithm which support the theoretical observations as well.

Intelligent recommender system based on quantum clustering and matrix completion

AbstractFaster than classical algorithms, quantum algorithms benefit from the superposition property of quantum information to offer significant speedup to complex algorithms. Therefore, quantum computing can be used to help machine learning algorithms by boosting their performance and accelerate the processing of time‐consuming ones. Clustering algorithms are very complex unsupervised learning algorithms. Indeed, the similarity calculation (distance) between input vectors is a resource‐consuming step, especially when working with large datasets. In this article, we propose a new better‐performing recommender system that operates as a combination of an adapted quantum K‐means algorithm and the singular value decomposition (SVT) algorithm. We integrate the developed quantum clustering algorithm to a prediction process of the proposed recommender system using matrix completion. To the best of our knowledge, no system with such details was proposed in the literature. The system was applied on the MovieLens dataset without a dimensionality reduction step and evaluated according to measures of information retrieval systems. The results of the quantum K‐means algorithm show that the quantum version leads to a logarithmic reduction of the time complexity compared to the classical algorithm. The proposed system has proved to be better than the previous tested ones in terms of precision and recall.

Automated detection of celiac disease using Machine Learning Algorithms

Celiac disease is a disorder of the immune system that mainly affects the small intestine but can also affect the skeletal system. The diagnosis relies on histological assessment of duodenal biopsies acquired by upper digestive endoscopy. Immunological tests involve collecting a blood sample to detect if the antibodies have been produced in the body. Endoscopy is invasive and histology is time-consuming. In recent years there have been various algorithms that use artificial intelligence (AI) and neural convolutions (CNN, Convolutional Neural Network) to process images from capsule endoscopy, a non-invasive endoscopy approach, that provides magnified, high qualitative images of the small bowel mucosa, to quickly establish a diagnosis. The proposed innovative approach do not use complex learning algorithms, instead it find some artefacts in the endoscopies using kernels and use classified machine learning algorithms. Each used artefacts have a psychical meaning: atrophies of the mucosa with a visible submucosal vascular pattern; the presence of cracks (depressions) that have an appearance similar to that of dry land; reduction or complete loss of folds in the duodenum; the presence of a submerged appearance at the Kerckring folds and a low number of villi. The results obtained for video capsule endoscopy images processing reveal an accuracy of 94.1% and F1 score of 94%, which is competitive with other complex algorithms. The main goal of the present research was to demonstrate that computer-aided diagnosis of celiac disease is possible even without the use of very complex algorithms, which require expensive hardware and a lot of processing time. The use of the proposed automated images processing acquired noninvasively by capsule endoscopy would be assistive in detecting the subtle presence of villous atrophy not evident by visual inspection. It may also be useful to assess the degree of improvement of celiac. Patients on a gluten-free diet, the main treatment method for stopping the autoimmune process and improving the state of the small intestinal villi. The novelty of the work is that the algorithm uses two modified filters to properly analyse the intestine wall texture. It is proved that using the right filters, the proper diagnostic can be obtained by image processing, without the use of a complicated machine learning algorithm.

Fusing compressed deep ConvNets with a self-normalizing residual block and alpha dropout for a cost-efficient classification and diagnosis of gastrointestinal tract diseases

The challenging task of diagnosing gastrointestinal (GI) tracts recently became a popular research topic, where most researchers performed extraordinary feats using numerous deep learning (DL) and computer vision techniques to achieve state-of-the-art (SOTA) diagnostic performance based on accuracy. However, most proposed methods relied on combining complex computational methods and algorithms, causing a significant increase in production difficulty, parameter size, and even training cost. Therefore, this method proposes a straightforward approach to developing a vision-based DL model without requiring heavy computing resources or reliance on other complex feature processing and learning algorithms. This paper included the step-by-step procedure consisting of network compression, layer-wise fusion, and the addition of a modified residual layer (MResBlock) with a self-normalizing attribute and a more robust regularization. In addition, the paper also presents the performance of the proposed method toward the diagnosis of four GI tract conditions, including polyps, ulcers, esophagitis, and healthy mucosa. The paper concludes that the proposed method did radiate a significant improvement in the overall performance, cost-efficiency, and especially practicality compared to most current SOTA methods.•The proposed method combined profound techniques like feature fusion, residual learning, and self-normalization to develop a lightweight model that accurately diagnoses gastrointestinal (GI) tract conditions.•The model produced from the proposed method generated better performance than most pre-existing state-of-the-art Deep Convolutional Neural Networks that diagnosed the presented four GI tract conditions.•Aside from its competitive performance, the model based on the proposed method only had 1.2M parameters and only consumed 1.5 GFLOPS, making it significantly more cost-efficient than most existing solutions.

Land-Use Classification via Transfer Learning with a Deep Convolutional Neural Network

Land cover classification provides efficient and accurate information regarding human land-use, which is crucial for monitoring urban development patterns, management of water and other natural resources, and land-use planning and regulation. However, land-use classification requires highly trained, complex learning algorithms for accurate classification. Current machine learning techniques already exist to provide accurate image recognition. This research paper develops an image-based land-use classifier using transfer learning with a pre-trained ResNet-18 convolutional neural network. Variations of the resulting approach were compared to show a direct relationship between training dataset size and epoch length to accuracy. Experiment results show that transfer learning is an effective way to create models to classify satellite images of land-use with a predictive performance. This approach would be beneficial to the monitoring and predicting of urban development patterns, management of water and other natural resources, and land-use planning.

Spatial-Temporal Hybrid Neural Network With Computing-in-Memory Architecture

Deep learning (DL) has gained unprecedented success in many real-world applications. However, DL poses difficulties for efficient hardware implementation due to the needs of a complex gradient-based learning algorithm and the required high memory bandwidth for synaptic weight storage, especially in today's data-intensive environment. Computing-in-memory (CIM) strategies have emerged as an alternative for realizing energy-efficient neuromorphic applications in silicon, reducing resources and energy required for neural computations. In this work, we exploit a CIM-based spatial-temporal hybrid neural network (STHNN) with a unique learning algorithm. To be specific, we integrate both multilayer perceptron and recurrent-based delay-dynamical system, making the network becomes linear separable while processing information in both spatial and temporal domains, better yet, reducing the memory bandwidth and hardware overhead through the CIM architecture. The prototype fabricated in 180 nm CMOS process is built of fully-analog components, yielding an average on-chip classification accuracy up to 86.9% on handprinted alphabet characters with a power consumption of 33 mW. Beyond that, through the handwritten digit database and the radio frequency fingerprinting dataset, software-based numerical evaluations offer 1.6 -to- 9.8 × and 1.9 -to- 4.4 × speedup, respectively, without significantly degrading its classification accuracy compared to the cutting-edge DL approaches.

Q-learning and LSTM based deep active learning strategy for malware defense in industrial IoT applications

Edge devices are extensively used as intermediaries between the device and the service layer in an industrial Internet of things (IIoT) environment. These devices are quite vulnerable to malware attacks. Existing studies have worked on designing complex learning algorithms or deep architectures to accurately classify malware assuming that a sufficient number of labeled examples are provided. In the real world, getting labeled examples is one of the major issues for training any classification algorithm. Recent advances have allowed researchers to use active learning strategies that are trained on a handful of labeled examples to perform the classification task, but they are based on the selection of informative instances. This study integrates the Q-learning characteristics into an active learning framework, which allows the network to either request or predict a label during the training process. We proposed the use of phase space embedding, sparse autoencoder, and LSTM with the action-value function to classify malware applications while using a handful of labeled examples. The network relies on its uncertainty to either request or predict a label. The experimental results show that the proposed method can achieve better accuracy than the supervised learning strategy while using few labeled requests. The results also show that the trained network is resilient to the adversarial attacks, which proves the robustness of the proposed method. Additionally, this study explores the tradeoff between classification accuracy and number of label requests via the choice of rewards and the use of decision-level fusion strategies to boost the classification performance. Furthermore, we also provide a hypothetical framework as an implication of the proposed method.

Machine Learning-based Flu Forecasting Study Using the Official Data from the Centers for Disease Control and Prevention and Twitter Data

Aim/Purpose: In the United States, the Centers for Disease Control and Prevention (CDC) tracks the disease activity using data collected from medical practice's on a weekly basis. Collection of data by CDC from medical practices on a weekly basis leads to a lag time of approximately 2 weeks before any viable action can be planned. The 2-week delay problem was addressed in the study by creating machine learning models to predict flu outbreak. Background: The 2-week delay problem was addressed in the study by correlation of the flu trends identified from Twitter data and official flu data from the Centers for Disease Control and Prevention (CDC) in combination with creating a machine learning model using both data sources to predict flu outbreak. Methodology: A quantitative correlational study was performed using a quasi-experimental design. Flu trends from the CDC portal and tweets with mention of flu and influenza from the state of Georgia were used over a period of 22 weeks from December 29, 2019 to May 30, 2020 for this study. Contribution: This research contributed to the body of knowledge by using a simple bag-of-word method for sentiment analysis followed by the combination of CDC and Twitter data to generate a flu prediction model with higher accuracy than using CDC data only. Findings: The study found that (a) there is no correlation between official flu data from CDC and tweets with mention of flu and (b) there is an improvement in the performance of a flu forecasting model based on a machine learning algorithm using both official flu data from CDC and tweets with mention of flu. Recommendations for Practitioners: In this study, it was found that there was no correlation between the official flu data from the CDC and the count of tweets with mention of flu, which is why tweets alone should be used with caution to predict a flu out-break. Based on the findings of this study, social media data can be used as an additional variable to improve the accuracy of flu prediction models. It is also found that fourth order polynomial and support vector regression models offered the best accuracy of flu prediction models. Recommendations for Researchers: Open-source data, such as Twitter feed, can be mined for useful intelligence benefiting society. Machine learning-based prediction models can be improved by adding open-source data to the primary data set. Impact on Society: Key implication of this study for practitioners in the field were to use social media postings to identify neighborhoods and geographic locations affected by seasonal outbreak, such as influenza, which would help reduce the spread of the disease and ultimately lead to containment. Based on the findings of this study, social media data will help health authorities in detecting seasonal outbreaks earlier than just using official CDC channels of disease and illness reporting from physicians and labs thus, empowering health officials to plan their responses swiftly and allocate their resources optimally for the most affected areas. Future Research: A future researcher could use more complex deep learning algorithms, such as Artificial Neural Networks and Recurrent Neural Networks, to evaluate the accuracy of flu outbreak prediction models as compared to the regression models used in this study. A future researcher could apply other sentiment analysis techniques, such as natural language processing and deep learning techniques, to identify context-sensitive emotion, concept extraction, and sarcasm detection for the identification of self-reporting flu tweets. A future researcher could expand the scope by continuously collecting tweets on a public cloud and applying big data applications, such as Hadoop and MapReduce, to perform predictions using several months of historical data or even years for a larger geographical area.

Molecule Sensitive Optical Imaging and Monitoring Techniques-A Review of Applications in Micro-Process Engineering.

This paper provides an overview of how molecule-sensitive, spatially-resolved technologies can be applied for monitoring and measuring in microchannels. The principles of elastic light scattering, fluorescence, near-infrared, mid-infrared, and Raman imaging, as well as combination techniques, are briefly presented, and their advantages and disadvantages are explained. With optical methods, images can be acquired both scanning and simultaneously as a complete image. Scanning technologies require more acquisition time, and fast moving processes are not easily observable. On the other hand, molecular selectivity is very high, especially in Raman and mid-infrared (MIR) scanning. For near-infrared (NIR) images, the entire measuring range can be simultaneously recorded with indium gallium arsenide (InGaAs) cameras. However, in this wavelength range, water is the dominant molecule, so it is sometimes necessary to use complex learning algorithms that increase the preparation effort before the actual measurement. These technologies excite molecular vibrations in a variety of ways, making these methods suitable for specific products. Besides measurements of the fluid composition, technologies for particle detection are of additional importance. With scattered light techniques and evaluation according to the Mie theory, particles in the range of 0.2–1 µm can be detected, and fast growth processes can be observed. Local multispectral measurements can also be carried out with fiber optic-coupled systems through small probe heads of approximately 1 mm diameter.

RETRACTED: A novel hybrid optimized and adaptive reconfigurable framework for the implementation of hybrid bio-inspired classifiers for diagnosis

Abstract Due to recent advances in IoT (Internet of Things) technologies, availability of reliable data and emergence of machine learning, bio-inspired learning and artificial intelligence, has demonstrated its ability to solve the large complex problems which is not possible before. In particular, machine learning and bio-inspired learning algorithms provides the effective solutions in image processing techniques. However, the implementation of the above-mentioned algorithms in the general CPU requires the intensive usage of bandwidth, area and power which makes the CPU unhealthy of usage and implementation. To overcome this problem, ASIC (application specific integrated circuits), GPU (Graphics Processing Unit) &FPGA (Field Programmable gate arrays) have been employed to improve the performance of the hybrid machine learning (ML) classifiers and deep learning algorithms. FPGA has been recently employed for an effective implementation and to achieve the high performance of the learning algorithms. But integrating the complex learning algorithms in FPGA still remains to be real challenge among the researchers. The paper proposes new reconfigurable architectures for bio- inspired classifiers to diagnosis the medical casualties which can be suitable for the tele health care applications. This paper aim is as follows (i) Design and implementation of Parallel Fusion of FSM and Reconfigurable shared Distributed Arithmetic for Bio-Inspired Classifiers (ii) Development of Accelerator Environment to test the performance of proposed architecture (iii) Performance evaluation of proposed architecture in terms of accuracy of detection in compared with MATLAB simulation iv) Implementation of proposed architectures in different ARtix-7 architectures and determination of power, throughput and area . Moreover, the proposed architecture has been tested with the and compared with the other existing architectures.

IMU Sensor-Based Hand Gesture Recognition for Human-Machine Interfaces.

We propose an efficient hand gesture recognition (HGR) algorithm, which can cope with time-dependent data from an inertial measurement unit (IMU) sensor and support real-time learning for various human-machine interface (HMI) applications. Although the data extracted from IMU sensors are time-dependent, most existing HGR algorithms do not consider this characteristic, which results in the degradation of recognition performance. Because the dynamic time warping (DTW) technique considers the time-dependent characteristic of IMU sensor data, the recognition performance of DTW-based algorithms is better than that of others. However, the DTW technique requires a very complex learning algorithm, which makes it difficult to support real-time learning. To solve this issue, the proposed HGR algorithm is based on a restricted column energy (RCE) neural network, which has a very simple learning scheme in which neurons are activated when necessary. By replacing the metric calculation of the RCE neural network with DTW distance, the proposed algorithm exhibits superior recognition performance for time-dependent sensor data while supporting real-time learning. Our verification results on a field-programmable gate array (FPGA)-based test platform show that the proposed HGR algorithm can achieve a recognition accuracy of 98.6% and supports real-time learning and recognition at an operating frequency of 150 MHz.

New method of training two-layer sigmoid neural networks using regularization

We propose a complex learning algorithm for sigmoid Artificial Neural Networks (ANN). We introduce the concept of the working area of a neuron for sigmoid ANNs in the form of a band in the attribute space, its width and location associated with the center line of the band to a fixed point. We define of the centers and widths of the working areas of neurons by analogy to the radial ANNs. On this basis, an algorithm for selecting the initial approximation of network parameters, ensuring uniform coverage of the data area with neuron working areas was developed. Network learning is carried out using a non-smooth regularizer designed to smooth and remove non-informative neurons. The results of the computational experiment illustrate the efficiency of the proposed integrated approach.