Single-layer Network Research Articles

Pitfalls of Using Multinomial Regression Analysis to Identify Class-Structure-Relevant Variables in Biomedical Data Sets: Why a Mixture of Experts (MOE) Approach Is Better

Recent advances in mathematical modeling and artificial intelligence have challenged the use of traditional regression analysis in biomedical research. This study examined artificial data sets and biomedical data sets from cancer research using binomial and multinomial logistic regression. The results were compared with those obtained with machine learning models such as random forest, support vector machine, Bayesian classifiers, k-nearest neighbors, and repeated incremental clipping (RIPPER). The alternative models often outperformed regression in accurately classifying new cases. Logistic regression had a structural problem similar to early single-layer neural networks, which limited its ability to identify variables with high statistical significance for reliable class assignments. Therefore, regression is not per se the best model for class prediction in biomedical data sets. The study emphasizes the importance of validating selected models and suggests that a “mixture of experts” approach may be a more advanced and effective strategy for analyzing biomedical data sets.

Research on multi-layer network routing selection strategy based on cooperative evolutionary game in IoT environment

As a new technology and application mode, the Internet of Things has an important impact on social life and economic development. In recent years, low-cost optimization of network transmission to solve the congestion problem of multi-layer communication networks has become one of the research hotspots. In this paper, a multi-dimensional communication data transmission model based on a multi-layer network is proposed. It then uses cooperative evolutionary game theory to calculate revenue, update weights, and adapt neighbors. Finally, the attention mechanism is dynamically introduced to share the weights of the multi-layer network, and the multi-dimensional communication propagation and routing strategies in the Internet of Things are studied and analyzed. The experimental results show that the model proposed in this paper has higher game revenue and application value than traditional single-layer network game theory. In particular, the indicators of cooperation rate, stable state, and maximum cooperation rate are better than the latter. The research results of this paper have important reference value for solving the problems of cooperation dilemma, social stickiness, and synergy in multi-layer networks.

A product requirement development method based on multi-layer heterogeneous networks

The rapid advancement of emerging technologies in the information age has resulted in intricate, diverse, and rapidly changing user requirements, thereby reducing the effectiveness of traditional requirement development approaches. This paper proposes a multi-layer heterogeneous network-based method for product requirement development to identify key requirements in the product development process and address the complex relationships among users, requirements, products, and requirement development teams. By incorporating the theory of hesitant fuzzy into the expert decision-making process, this method constructs four single-layer networks: user attention network, expert trust network, requirement correlation network, and component combination network, which assigns realistic significance to nodes in each network layer and establishes connections between different networks, forming a multi-layer heterogeneous network model for product requirement development. Furthermore, this paper defines indicators such as layer significance and node comprehensive significance and proposes a calculation method for node comprehensive significance based on biased random walk, which enables deep integration of node’s realistic significance and topological centrality. The proposed method is applied to a data experiment of product requirement development for a specific type of new energy vehicle. By analyzing the calculation results of node comprehensive significance, high-quality recommendations are provided for the development team from the perspectives of functional requirements, product components, and user groups. Moreover, comparative experiments to assess algorithm performance demonstrate that the average accuracy of the proposed algorithm is 0.95, which is significantly higher than that of the commonly used single-layer network node centrality measurement methods and recently proposed multi-layer network node centrality measurement methods. These results reflect the advancement and applicability of the proposed method in product development scenarios.

Cross-subject EEG linear domain adaption based on batch normalization and depthwise convolutional neural network

Electroencephalogram (EEG)-based emotion recognition has been widely used in affective computing. However, the study on improving recognition accuracy across individuals is insufficient. In this study, a new linear domain adaption approach with experiment-level batch normalization and a single-layer depthwise convolutional neural network is proposed. In particular, the experiment-level batch normalization and depthwise convolutional neural network can be integrated as a linear mapping with a scaling parameter and a translation parameter. By linear mapping, difference between subjects in different domain can be effectively diminished, and the mapping parameters can be used to further investigate EEG emotion mechanism. The domain adaption experiments are conducted with SJTU emotion EEG dataset and SJTU emotion EEG dataset-IV, which are divided into source domain and target domain to validate the recognition effect across individuals. Multiple traditional machine learning and deep learning classifiers are used to examine the effectiveness of the proposed approach. By mapping the EEG data from source domain to target domain, the increment of recognition accuracy is up to 61.11% when using the support vector machine classifier. The highest recognition accuracy 97.22% is achieved when using the logistic regression classifier. The scaling and translation parameters in the mapping procedure are then analyzed with statistical methods. It is found that EEG signal waves in the same emotion category are highly similar and EEG data have characteristics including integration of channels and hierarchy of frequency bands. In addition, the experimental results indicate that emotion complexity and emotion sensitiveness of brain cortex regions can affect the correlations between channels.

An intelligent encryption decision method for autonomous domain of multilayer satellite network

Multi-layer satellite network (MLSN) is an important part of the Space-Air-Ground Integrated Network (SAGIN), it has obvious advantages in coverage density, business processing, service efficiency and other comprehensive capabilities compared with ordinary single-layer satellite networks. However, due to the fact that many space nodes, highly dynamic topology changes and wireless links scattered in free space, MLSNs are facing the dual challenges of inefficient organization and lack of security. In this case, this paper considers providing secure encryption decision support for the return of data obtained by low earth orbit (LEO) observation satellites relying on the local autonomous region of multi-layer satellite space. By quantifying the security strength of encryption algorithms, the work to be optimized is described as the problem of maximizing the security strength under the constraint of return delay of the data. This paper thus proposes a solution based on deep reinforcement learning (DRL), which relies on reasonably setting the key supporting elements such as state, action and reward closely related to the problem to achieve fast and reliable decision output, and the simulation results show that the proposed method has better performance than many comparison methods.

Technology opportunity analysis using hierarchical semantic networks and dual link prediction

Technology opportunity analysis using network analysis and link prediction has attracted the interest of both academia and industry. However, there are several unresolved issues with existing research, such as a lack of semantic relationships between nodes in a single-layer network, analyzing current technology trends rather than predicting future technology developments based on existing edges in a semantic network, and ignoring evaluation criteria for technology opportunity identification based on a single link prediction. This study proposes a new systematic methodology to address these issues and identify technology opportunities using a hierarchical semantic network and dual link prediction. The proposed methodology consists of three modules: 1) constructing the hierarchical semantic network based on SAO structures extracted from patents; 2) identifying technology opportunities in this semantic network through probabilistic-based link prediction; and 3) evaluating these opportunities via similarity-based link prediction. The viability and usefulness of the proposed methodology is proved by empirical analysis of the exploitation technology in the coal seam gas (CSG) industry. The results show that the hierarchical semantic network, including semantic and co-word relationships, can improve prediction accuracy. The dual link prediction can not only automatically identify technology opportunities with semantics, but also evaluate them to narrow down the problem-solving according to the novelty criteria. This study represents a contribution to existing research on technology opportunity analysis by integrating the hierarchical semantic network and dual link prediction.

Enhancing multiplex global efficiency

Modeling complex systems that consist of different types of objects leads to multilayer networks, in which vertices are connected by both inter-layer and intra-layer edges. In this paper, we investigate multiplex networks, in which vertices in different layers are identified with each other, and the only inter-layer edges are those that connect a vertex with its copy in other layers. Let the third-order adjacency tensor A∈RN×N×L\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\mathcal {A}\\in \\mathbb {R}^{N\ imes N\ imes L}$$\\end{document} and the parameter γ≥0\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\gamma \\ge 0$$\\end{document}, which is associated with the ease of communication between layers, represent a multiplex network with N vertices and L layers. To measure the ease of communication in a multiplex network, we focus on the average inverse geodesic length, which we refer to as the multiplex global efficiency eA(γ)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$e_\\mathcal {A}(\\gamma )$$\\end{document} by means of the multiplex path length matrix P∈RN×N\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$P\\in \\mathbb {R}^{N\ imes N}$$\\end{document}. This paper generalizes the approach proposed in [15] for single-layer networks. We describe an algorithm based on min-plus matrix multiplication to construct P, as well as variants PK\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$P^K$$\\end{document} that only take into account multiplex paths made up of at most K intra-layer edges. These matrices are applied to detect redundant edges and to determine non-decreasing lower bounds eAK(γ)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$e_\\mathcal {A}^K(\\gamma )$$\\end{document} for eA(γ)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$e_\\mathcal {A}(\\gamma )$$\\end{document}, for K=1,2,⋯,N-2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$K=1,2,\\dots ,N-2$$\\end{document}. Finally, the sensitivity of eAK(γ)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$e_\\mathcal {A}^K(\\gamma )$$\\end{document} to changes of the entries of the adjacency tensor A\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\mathcal {A}$$\\end{document} is investigated to determine edges that should be strengthened to enhance the multiplex global efficiency the most.

Predicting the BRAF mutation with pretreatment MRI radiomics features for melanoma brain metastases receiving Gamma Knife radiosurgery

To develop a model using radiomics features extracted from magnetic resonance imaging (MRI) images of Gamma Knife radiosurgery (GKRS) to predict the BRAF mutation in patients with melanoma brain metastases (MBM). Data of 220 tumours were classified into two groups. One was a group whose BRAF mutation was identified, and the other group whose BRAF mutation was not identified. We extracted 1,962 radiomics features from gadolinium contrast-enhanced T1-weighted MRI treatment-planning images. Synthetic Minority Over-sampling TEchnique (SMOTE) was performed to address the unbalanced data-related issues. A single-layer neural network (NN) was used to build predictive models with radiomics features. The sensitivity, specificity, accuracy, and the area under the curve (AUC) were evaluated to assess the model performance. The prediction performance for the final evaluation without the SMOTE had an accuracy of 77.14%, a specificity of 82.44%, a sensitivity of 81.85%, and an AUC of 0.79. The application of SMOTE improved the prediction model to an accuracy of 83.1%, a specificity of 87.07%, a sensitivity of 78.82%, and an AUC of 0.82. The current study showed the feasibility of generating a highly accurate NN model for the BRAF mutation prediction. The prediction performance improved with SMOTE. The model assists physicians to obtain more accurate expectations of the treatment outcome without a genetic test.

MARKOV NETWORK MODEL

Markov networks synthesized from the N-scheme of the Markov algorithm, developed to replace his well-known y-scheme, are studied theoretically from the standpoint of theories of algorithms and categories. The provisions of the theory of Markov algorithms are developed in the direction of synthesizing methods for the joint processing of words and morphism values that are occurrences in these words. In Markov networks, morphisms are weighted according to the method adopted in artificial neural networks. The material for the synthesis of Markov networks are diagrams of Markov occurrences of words into each other, obtained from input data samples, by classifying occurrences in the alphabet M = λιτφпо1…оx/αβδγχκμ. The Markov network model and the activation function of its morphemes are synthesized and subjected to scientific analysis. The idea of joint consideration of category theory and the theory of Markov algorithms is developed for the control channel, so this control channel is understood as an n-category associator introduced into the N-scheme of the Markov algorithm. In Markov networks, chain termination mechanisms are studied, using the control channel command, using combined methods of classifying words and morphisms, previously studied for the N-scheme of his algorithm, the evolution of categories formed by words is considered due to their occurrences in each other. For the method of synthesizing training samples from diagrams of Markov occurrences, a scientific substantiation of prescriptions for processing samples is carried out. For activation functions, examples of their implementation are considered. The categorical representation of the activation function is synthesized for Markov networks and subjected to a separate scientific analysis. For single-layer and multi-layer Markov networks of direct distribution, a block diagram has been developed. In Markov networks, the concept of zero-nets is introduced. As a promising direction for further theoretical research in Markov networks, multilayer Markov networks of indirect propagation are singled out, which have a conclusion that cannot be considered normal according to Markov. In the conclusions to the study, it is noted that Markov networks have a category-theoretic representation and can be synthesized from Markov zero-nets.

Analytical and Numerical Study of Information Retrieval Method Based on Single-Layer Neural Network with Optimization of Computing Algorithm Performance

This work presents a mathematical model of a fast-acting single-layer artificial neural network applied to the task of image reconstruction after noise. For research purposes, this algorithm was implemented in the Python and C++ programming languages. The numerical simulation of the recovery efficiency of the described neural network was performed for different values of the noise factor, the number of samples required to train elements in the sample and the dimensionality of the coupling coefficients, w. The study of the mathematical model of this neural network is presented; as a result, it is possible to identify its essence, to reduce the number of operations required to recover a single element and to increase recovery accuracy by changing the order of calculation of coupling coefficients, w.

A study on community detection in multilayer networks

Studying complex networks is essential for a better understanding of network science. Many studies have been done on single-layer networks in complex networks. After the advancement and widespread usage of the internet and social media networks, performing community detection in multilayer networks becomes essential to reach more people and work with different personalities on different platforms. Motivated by this observation, this paper has studied types of networks, metrics, measures, and community detection using deep learning-based models in multilayer networks. This survey can play a significant role in analyzing and understanding multilayer networks.

Analysis of Weather Factors on Aircraft Cancellation using a Multilayer Complex Network.

Airlines provide one of the most popular and important transportation services for passengers. While the importance of the airline industry is rising, flight cancellations are also increasing due to abnormal weather factors, such as rainfall and wind speed. Although previous studies on cancellations due to weather factors considered both aircraft and weather factors concurrently, the complex network studies only treated the aircraft factor with a single-layer network. Therefore, the aim of this study was to apply a multilayer complex network (MCN) method that incorporated three different factors, namely, aircraft, rainfall, and wind speed, to investigate aircraft cancellations at 14 airports in the Republic of Korea. The results showed that rainfall had a greater impact on aircraft cancellations compared with wind speed. To find out the most important node in the cancellation, we applied centrality analysis based on information entropy. According to the centrality analysis, Jeju Airport was identified as the most influential node since it has a high demand for aircraft. Also, we showed that characteristics and factors of aircraft cancellation should be appropriately defined by links in the MCN. Furthermore, we verified the applicability of the MCN method in the fields of aviation and meteorology. It is expected that the suggested methodology in this study can help to understand aircraft cancellation due to weather factors.

PREDICTING CONSTRUCTION CREW PRODUCTIVITY FOR CONCRETE POURING OPERATIONS OF RC COLUMNS USING ANN

In current practice, construction crew productivity (CCP) assessment solely relies on traditional methods like personal judgements or past published data. However, a systematic approach is required to measure and predict the CCP. Modelling with artificial neural networks (ANNs) can be useful for examining interrelationships between factors affecting CCP. The present study aims to predict CCP for concrete pouring operation of reinforced concrete (RC) columns using ANN. To achieve this, 14 factors like crew size, average age of crew, average experience of crew, extent of supervision, working hours, lead distance, working height, location of columns, number of concrete bucket carriers, method of concrete supply, temperature, wind speed, column length, and column width were identified from literature study and industry experts. These identified factors were considered as inputs, while CCP was the output variable for ANN model. The study collected 187 samples from 11 residential and commercial construction sites located in four cities of India. A single-layer feed-forward back propagation neural network was used for prediction. The developed model can estimate productivity rates reasonably with least range of errors. Outcomes of study can be useful to improve the crew’s productivity at construction sites for concrete pouring works of RC columns.

Extreme learning machine-based fault location approach for terminal-hybrid LCC-VSC-HVDC transmission lines

This paper proposes an Extreme Learning Machine-based fault location method to evaluate the DC transmission line faults of Terminal-hybrid LCC-VSC-HVDC systems. For this purpose, the voltage and current fault signals are firstly measured from the terminals of the hybrid HVDC system. Then, the high-frequency components of the input signals will be extracted and decomposed employing wavelet transform. In the next step, the energy of these extracted signals is calculated and provided to the extreme learning machine as intake train and test data. Finally, the exact location of the fault point will be determined using ELM. The proposed strategy is also applied to the same hybrid system using single-layer and double-layer neural networks, and the results have been thoroughly compared with ELM. The outcomes unequivocally indicate that by increasing the amount of DC filters in the transmission line, the average percentage error of fault location will be remarkably diminished. A ± 100 kV Terminal-hybrid LCC-VSC-HVDC system has been implemented and simulated via MATLAB software to assess the proposed method. It can be perceived that the ELM-based method not only works properly under various fault resistances, fault locations, and types of faults but also has significant superiority over the ANN-based approaches.

Percolation phase transition in the heterogeneous multi-coupled interdependent network

In previous studies, complex system models have often been set up in an oversimplified manner and the failure mechanism for nodes affected by each other is not correspond to reality. Accordingly, we develop a new model of heterogeneous multi-coupled interdependent networks and analyze the change in the giant component of the network after the attack in conjunction with the load redistribution mechanism. We first analyze the cascading failure process of the network analytically by recursive methods, estimate the critical phase transition threshold and the size of the giant component of the remaining network after network percolation, and find that the threshold of the WS sub-network is positively related to the average degree, and the BA sub-network is related to the power exponent λ of the distribution. When λ is less than or equal to 2, there is always a giant component in the network. After λ is greater than 3.4788⋯, there is no giant component. When λ is between 2 to 3.4788⋯, we found that the threshold of the sub-network is more influenced by the proportion of attacks (p1 and p2), and when p1 is larger, the sub-networks interact with each other, resulting in a smaller threshold for the BA sub-network, while the threshold of the WS sub-network is smaller in other cases. Finally, the cross-sectional comparison illustrates that as λ increases to 3.4788⋯, the robustness of the BA sub-network increases, contrary to the findings for single-layer networks. Overall, the heterogeneous multi-coupled interdependent network is more robust, and the giant component is greatly affected by the low robust sub-network, but its existence is related to the stock of nodes.

Pattern Formation in a Predator–Prey Model with Allee Effect and Hyperbolic Mortality on Multiplex Networks

With the rapid development of network science, Turing patterns on complex networks have attracted extensive attention from researchers. In this paper, we focus on spatial patterns in multiplex ER (Erdös-Rényi) random networks, taking the predator–prey model with Allee effect and hyperbolic mortality as an example. In theory, the threshold condition for generating Turing patterns is given using the Turing instability theory of multiplex networks. Numerically, we design relevant experiments to explore the impact of network topology on Turing patterns. The factors considered include model parameters, diffusion rate, average degree of the network, and differences in the average degree of different layers. The results indicate that the importance of diffusion rate and network average degree for Turing patterns is affirmed on the single-layer network. For multiplex networks, the differentiation of average degrees in different layers controls the generation of Turing patterns, which are not affected by the diffusion rates of the two populations. More interestingly, we observe the switching of Turing patterns and spatiotemporal patterns. We believe that these findings contribute to a better understanding of self-organization on complex networks.

Statistical mechanics of continual learning: Variational principle and mean-field potential.

An obstacle to artificial general intelligence is set by continual learning of multiple tasks of a different nature. Recently, various heuristic tricks, both from machine learning and from neuroscience angles, were proposed, but they lack a unified theory foundation. Here, we focus on continual learning in single-layered and multilayered neural networks of binary weights. A variational Bayesian learning setting is thus proposed in which the neural networks are trained in a field-space, rather than a gradient-ill-defined discrete-weight space, and furthermore, weight uncertainty is naturally incorporated, and it modulates synaptic resources among tasks. From a physics perspective, we translate variational continual learning into a Franz-Parisi thermodynamic potential framework, where previous task knowledge serves as a prior probability and a reference as well. We thus interpret the continual learning of the binary perceptron in a teacher-student setting as a Franz-Parisi potential computation. The learning performance can then be analytically studied with mean-field order parameters, whose predictions coincide with numerical experiments using stochastic gradient descent methods. Based on the variational principle and Gaussian field approximation of internal preactivations in hidden layers, we also derive the learning algorithm considering weight uncertainty, which solves the continual learning with binary weights using multilayered neural networks, and performs better than the currently available metaplasticity algorithm in which binary synapses bear hidden continuous states and the synaptic plasticity is modulated by a heuristic regularization function. Our proposed principled frameworks also connect to elastic weight consolidation, weight-uncertainty modulated learning, and neuroscience-inspired metaplasticity, providing a theoretically grounded method for real-world multitask learning with deep networks.

Global stock markets risk contagion: Evidence from multilayer connectedness networks in the frequency domain

Multilayer connectedness networks are a promising tool for unveiling the contagion mechanism of financial risk. This paper constructs multilayer connectedness networks in the frequency domain to examine the risk contagion among global stock markets from January 4, 2006, to June 30, 2022. We analyze the global efficiency, average connectedness strength, and network density of single-layer connectedness network, and investigate the average overlap degree, network correlation coefficient and network participation coefficients of inter-layer connectedness networks. We observe that (i) the risk contagion among global stock markets shows different behaviors in the short-, medium-, and long-term, (ii) during the period of financial stress, the medium- and long-term connectedness increase significantly, while the short-term connectedness decrease significantly, and (iii) the Asian stock markets are main risk receivers, but the risks they receive are heterogeneous in frequency. Our work provides a new perspective for studying global risk contagion and supply valuable knowledge for investor and regulators.

GPU-Free Specific Emitter Identification Using Signal Feature Embedded Broad Learning

Emerging wireless networks may suffer severe security threats due to the ubiquitous access of massive wireless devices. Specific emitter identification (SEI) is considered as one of the important techniques to protect wireless networks, which aims to identifying legal or illegal devices through the radio frequency (RF) fingerprints contained in RF signals. Existing SEI methods are implemented with either traditional machine learning or deep learning. The former relies on manual feature extraction which is usually inefficient, while the latter relies on the powerful GPU computing power but with limited applications and high cost. To solve these problems, in this paper, we propose a GPU-free SEI method using a signal feature embedded broad learning network (SFEBLN), for efficient emitter identification based on a single-layer forward propagation network on the CPU platform. With this method, the original RF data is first pre-processed through external signal processing nodes, and then processed to generate mapped feature nodes and enhancement nodes by nonlinear transformation. Next, we design the internal signal processing nodes to extract effective features from the processed RF signals. The final input layer consists of mapped feature nodes, enhancement nodes and internal signal processing nodes. Then, the network weight parameters are obtained by solving the pseudo inverse problem. Experiments are conducted over the CPU platform and the results show that our proposed SEI method using SFEBLN achieves a superior identification performance and robustness under various scenarios.

Lightweight Football Motion Recognition and Intensity Analysis Using Low-Cost Wearable Sensors.

In recent years, machine learning has been utilized in health informatics and sports science. There is a great demand and development potential for combining the Internet of Things (IoT) and artificial intelligence (AI) to be applied to football sports. The conventional teaching and training methods of football sports have limited collection and mining of real raw data using wearable devices, and lack human motion capture and gesture recognition based on sports science theories. In this study, a low-cost AI + IoT system framework is designed to recognize football motion and analyze motion intensity. To reduce the communication delay and the computational resource consumption caused by data operations, a multitask learning model is designed to achieve motion recognition and intensity estimation. The model can perform classification and regression tasks in parallel and output the results simultaneously. A feature extraction scheme is designed in the initial data processing, and feature data augmentation is performed to solve the small sample data problem. To evaluate the performance of the designed football motion recognition algorithm, this paper proposes a data extraction experimental scheme to complete the data collection of different motions. Model validation is performed using three publicly available datasets, and the features learning strategies are analyzed. Finally, experiments are conducted on the collected football motion datasets and the experimental results show that the designed multitask model can perform two tasks simultaneously and can achieve high computational efficiency. The multitasking single-layer long short-term memory (LSTM) network with 32 neural units can achieve the accuracy of 0.8372, F1 score of 0.8172, mean average precision (mAP) of 0.7627, and mean absolute error (MAE) of 0.6117, while the multitasking single-layer LSTM network with 64 neural units can achieve the accuracy of 0.8407, F1 score of 0.8132, mAP of 0.7728, and MAE of 0.5966.