A graph neural network approach to the inverse design for thermal transparency with periodic interparticle system

Recent years have witnessed significant advances in utilizing machine learning-based techniques for thermal metamaterial-based structures and devices to attain favorable thermal transport behaviors.Among the various thermal transport behaviors,achieving thermal transparency stands out as particularly desirable and intriguing.Our earlier work demonstrated the use of a thermal metamaterial-based periodic interparticle system as the underlying structure for manipulating thermal transport behavior and achieving thermal transparency.In this paper,we introduce an approach based on graph neural network to address the complex inverse design problem of determining the design parameters for a thermal metamaterial-based periodic interparticle system with the desired thermal transport behavior.Our work demonstrates that combining graph neural network modeling and inference is an effective approach for solving inverse design problems associated with attaining desirable thermal transport behaviors using thermal metamaterials.

Cooperative Content Caching and Delivery in Vehicular Networks: A Deep Neural Network Approach

The growing demand for low delay vehicular content has put tremendous strain on the backbone network.As a promising alternative,cooperative content caching among different cache nodes can reduce content access delay.However,heterogeneous cache nodes have different communication modes and limited caching capacities.In addition,the high mobility of vehicles renders the more complicated caching environment.Therefore,performing efficient cooperative caching becomes a key issue.In this paper,we propose a cross-tier cooperative caching architecture for all contents,which allows the distributed cache nodes to cooperate.Then,we devise the communication link and content caching model to facilitate timely content delivery.Aiming at minimizing transmission delay and cache cost,an optimization problem is formulated.Furthermore,we use a multi-agent deep reinforcement learning(MADRL)approach to model the decision-making process for caching among heterogeneous cache nodes,where each agent interacts with the environment collectively,receives observations yet a common reward,and learns its own optimal policy.Extensive simulations validate that the MADRL approach can enhance hit ratio while reducing transmission delay and cache cost.

Residential Community Open-Up Strategy Based on Prim’s Algorithm and Neural Network Algorithm

“Open community” has aroused widespread concern and research. This paper focuses on the system analysis research of the problem that based on statistics including the regression equation fitting function and mathematical theory, combined with the actual effect of camera measurement method, Prim’s algorithm and neural network to “Open community” and the applicable conditions. Research results show that with the increasing number of roads within the district, the benefit time gradually increased, but each type of district capacity is different.

Prediction Method for Network Traffic Based on Maximum Correntropy Criterion

This paper proposes a method for improving the precision of Network Traffic Prediction based on the Maximum Correntropy Criterion(NTPMCC),where the nonlinear characteristics of network traffic are considered.This method utilizes the MCC as a new error evaluation criterion or named the cost function(CF)to train neural networks(NN).MCC is based on a new similarity function(Generalized correlation entropy function,Correntropy),which has as its foundation the Parzen window evaluation and Renyi entropy of error probability density function.At the same time,by combining the MCC with the Mean Square Error(MSE),a mixed evaluation criterion with MCC and MSE is proposed as a cost function of NN training.According to the traffic network characteristics including the nonlinear,non-Gaussian,and mutation,the Elman neural network is trained by MCC and MCC-MSE,and then the trained neural network is used as the model for predicting network traffic.The simulation results based on the evaluation by Mean Absolute Error(MAE),MSE,and Sum Squared Error(SSE)show that the accuracy of the prediction based on MCC is superior to the results of the Elman neural network with MSE.The overall performance is improved by about 0.0131.

Incremental Learning of Radio Modulation Classification Based on Sample Recall

Radio modulation classification has always been an important technology in the field of communications.The difficulty of incremental learning in radio modulation classification is that learning new tasks will lead to catastrophic forgetting of old tasks.In this paper,we propose a sample memory and recall framework for incremental learning of radio modulation classification.For data with different signal-to-noise ratios,we use a partial memory strategy by selecting appropriate samples for memorizing.We compare the performance of our proposed method with three baselines through a large number of simulation experiments.Results show that our method achieves far higher classification accuracy than finetuning method and feature extraction method.Furthermore,it performs closely to joint training method which uses all old data in terms of classification accuracy which validates the effectiveness of our method against catastrophic forgetting.

Constructive Approximation by Superposition of Sigmoidal Functions

In this paper, a constructive theory is developed for approximating func- tions of one or more variables by superposition of sigmoidal functions. This is done in the uniform norm as well as in the L^p norm. Results for the simultaneous approx- imation, with the same order of accuracy, of a function and its derivatives （whenever these exist）, are obtained. The relation with neural networks and radial basis func- tions approximations is discussed. Numerical examples are given for the purpose of illustration.

Environmental Sound Event Detection in Wireless Acoustic Sensor Networks for Home Telemonitoring

In this paper, we present an approach to improve the accuracy of environmental sound event detection in a wireless acoustic sensor network for home monitoring. Wireless acoustic sensor nodes can capture sounds in the home and simultaneously deliver them to a sink node for sound event detection. The proposed approach is mainly composed of three modules, including signal estimation, reliable sensor channel selection, and sound event detection. During signal estimation, lost packets are recovered to improve the signal quality. Next, reliable channels are selected using a multi-channel cross-correlation coefficient to improve the computational efficiency for distant sound event detection without sacrificing performance. Finally, the signals of the selected two channels are used for environmental sound event detection based on bidirectional gated recurrent neural networks using two-channel audio features. Experiments show that the proposed approach achieves superior performances compared to the baseline.

Color space lip segmentation for drivers＇ fatigue detection

to the chroma distribution diversity （CDD） between lip color and skin color, the lip color area is segmented by the back propagation neural network （BPNN） with three typical color features. Isolated noisy points of the lip color area in binary image are eliminated by a proposed re- gion connecting algorithm. An improved integral projection algorithm is presented to locate the lip boundary. Whether a driver is fatigued is recognized by the ratio of the frame number of the images with mouth opening continuously to the total image frame number in every 20s. The experiments show that the proposed algorithm provides higher correct rate and reliability for fatigue driving detec- tion, and is superior to the single color feature-based method in the lip color segmention. Besides, it improves obviously the accuracy and speed of the lip boundary location compared with the traditional integral projection algrothm.

Deep-learning-based methods for super-resolution fluorescence microscopy

The algorithm used for reconstruction or resolution enhancement is one of the factors affectingthe quality of super-resolution images obtained by fluorescence microscopy.Deep-learning-basedalgorithms have achieved stateof-the-art performance in super-resolution fluorescence micros-copy and are becoming increasingly attractive.We firstly introduce commonly-used deep learningmodels,and then review the latest applications in terms of the net work architectures,the trainingdata and the loss functions.Additionally,we discuss the challenges and limits when using deeplearning to analyze the fluorescence microscopic data,and suggest ways to improve the reliability and robustness of deep learning applications.

A hybrid model of a subminiature helicopter in horizontal turn

A hybrid model of a subminiature helicopter in horizontal turn is presented. This model is based on a mechanism model and its compensated neural network （NN）. First, the nonlinear dynamics of a sub-miniature helicopter is established. Through the linearization of the nonlinear dynamics on a trim point, the linear time-invariant mechanism model in horizontal turn is obtained. Then a diagonal recursive neural network is used to compensate the model error between the mechanism model and the nonlinear model, thus the hybrid model of a subminiature helicopter in horizontal turn is achieved. Simulation results show that the hybrid model has higher accuracy than the mechanism model and the obtained compensated-NN has good generalization capability.

A Novel Method for Solving Nonlinear Schrödinger Equation with a Potential by Deep Learning

The improved physical information neural network algorithm has been proven to be used to study integrable systems. In this paper, the improved physical information neural network algorithm is used to study the defocusing nonlinear Schrödinger (NLS) equation with time-varying potential, and the rogue wave solution of the equation is obtained. At the same time, the influence of the number of network layers, neurons and the number of sampling points on the network performance is studied. Experiments show that the number of hidden layers and the number of neurons in each hidden layer affect the relative L2-norm error. With fixed configuration points, the relative norm error does not decrease with the increase in the number of boundary data points, which indicates that in this case, the number of boundary data points has no obvious influence on the error. Through the experiment, the rogue wave solution of the defocusing NLS equation is successfully captured by IPINN method for the first time. The experimental results of this paper are also compared with the results obtained by the physical information neural network method and show that the improved algorithm has higher accuracy. The results of this paper will be contributed to the generalization of deep learning algorithms for solving defocusing NLS equations with time-varying potential.

Drogue detection for autonomous aerial refueling based on convolutional neural networks

Drogue detection is a fundamental issue during the close docking phase of autonomous aerial refueling（AAR）. To cope with this issue, a novel and effective method based on deep learning with convolutional neural networks（CNNs） is proposed. In order to ensure its robustness and wide application, a deep learning dataset of images was prepared by utilizing real data of ‘‘Probe and Drogue＂ aerial refueling, which contains diverse drogues in various environmental conditions without artificial features placed on the drogues. By employing deep learning ideas and graphics processing units（GPUs）, a model for drogue detection using a Caffe deep learning framework with CNNs was designed to ensure the method＇s accuracy and real-time performance. Experiments were conducted to demonstrate the effectiveness of the proposed method, and results based on real AAR data compare its performance to other methods, validating the accuracy, speed, and robustness of its drogue detection ability.

Research on axial bearing capacity of rectangular concrete-filled steel tubular columns based on artificial neural networks

Design of rectangular concrete-filled steel tubular （CFT） columns has been a big concern owing to their complex constraint mechanism. Generally, most existing methods are based on simplified mechanical model with limited experimental data, which is not reliable under many conditions, e.g., columns using high strength materials. Artificial neural network （ANN） models have shown the effectiveness to solve complex problems in many areas of civil engineering in recent years. In this paper, ANN models were employed to predict the axial bearing capacity of rectangular CFT columns based on the experimental data. 305 experimental data from articles were collected, and 275 experimental samples were chosen to train the ANN models while 30 experimental samples were used for testing. Based on the comparison among different models, artificial neural network modell （ANN1） and artificial neural network model2 （ANN2） with a 20- neuron hidden layer were chosen as the fit prediction models. ANN1 has five inputs： the length （D） and width （B） of cross section, the thickness of steel （t）, the yield strength of steel （fy）, the cylinder strength of concrete （fc＇）- ANN2 has ten inputs： D, B, t, fy, f′, the length to width ratio （D/B）, the length to thickness ratio （D/t）, the width to thickness ratio （B/t）, restraint coefficient （ξ）, the steel ratio （α）. The axial beating capacity is the output data for both models.The outputs from ANN1 and ANN2 were verified and compared with those from EC4, ACI, GJB4142 and AISC360-10. The results show that the implemented models have good prediction and generalization capacity. Parametric study was conducted using ANN1 and ANN2 which indicates that effect law of basic parameters of columns on the axial bearing capacity of rectangular CFT columns differs from design codes.The results also provide convincing design reference to rectangular CFT columns.

Recognition of Continuous Digits by Quantum Neural Networks

This paper describes a new kind of neural network-Quantum Neural Network(QNN) and its application to recognition of continuous digits. QNN combines the advantages of neuralmodeling and fuzzy theoretic principles . Experiment results show that more than 15 percent errorreduction is achieved on a speaker-independent continuous digits recognition task compared with BPnetworks.

Correlation Analysis for the Attack of Bacillary Dysentery and Meteorological Factors Based on the Chinese Medicine Theory of Yunqi and the Medical-Meteorological Forecast Model

Objective： To explore the impact of meteorological factors on the outbreak of bacillary dysentery, so as to provide suggestions for disease prevention. Methods： Based on the Chinese medicine theory of Yunqi, the descriptive statistics, single-factor correlation analysis and back-propagation artificial neural net-work were conducted using data on five basic meteorological factors and data on incidence of bacillary dysentery in Beijing, China, for the period 1970-2004. Results： The incidence of bacillary dysentery showed significant positive correlation relationship with the precipitation, relative humidity, vapor pressure, and temperature, respectively. The incidence of bacillary dysentery showed a negatively correlated relationship with the wind speed and the change trend of average wind speed. The results of medical-meteorological forecast model showed a relatively high accuracy rate. Conclusions： There is a close relationship between the meteorological factors and the incidence of bacillary dysentery, but the contributions of which to the onset of bacillary dysentery are different to each other.

A Fully-Decoupled RAN Architecture for 6G Inspired by Neurotransmission

While the commercial deployment and promotion of 5G is ongoing,mobile communication networks are still facing three fundamental challenges,i.e.,spectrum resource scarcity,especially for low-frequency spectrum,exacerbated by fragmented spectrum allocation,user-centric network service provision when facing billions of personalized user demands in the era of Internet of everything(IoE),and proliferating operation costs mainly due to huge energy consumption of network infrastructure.To address these issues,it is imperative to consider and develop disruptive technologies in the next generation mobile communication networks,namely 6G.In this paper,by studying brain neurons and the neurotransmission,we propose the fully-decoupled radio access network(FD-RAN).In the FD-RAN,base stations(BSs)are physically decoupled into control BSs and data BSs,and the data BSs are further physically split into uplink BSs and downlink BSs.We first review the fundamentals of neurotransmission and then propose the 6G design principles inspired by the neurotransmission.Based on the principles,we propose the FD-RAN architecture,elastic resource cooperation in FD-RAN,and improved transport service layer design.The proposed fully decoupled and flexible architecture can profoundly facilitate resource cooperation to enhance the spectrum utilization,reduce the network energy consumption and improve the quality of user experience.Future research topics in this direction are envisioned and discussed.

Adaptive SRM neuron based on NbO_(x) memristive device for neuromorphic computing

The spike-response model(SRM)describes the adaptive behaviors of a biological neuron in response to repeated or prolonged stimulation,so that SRM neurons can avoid information overload and support neural networks for competitive learning.In this work,an artificial SRM neuron with the leaky integrate-and-fire(LIF)functions and the adaptive threshold is firstly implemented by the volatile memris-tive device of Pt/NbO_(x)/TiN.By modulating the volatile speed of the device,the threshold of the SRM neuron is adjusted to achieve the adaptive behaviors,such as the refractory period and the lateral inhi-bition.To demonstrate the function of the SRM neuron,a spiking neu-ral network(SNN)is constructed with the SRM neurons and trained by the unsupervised learning rule,which successfully classifies letters with noises,while a similar SNN with LIF neurons fails.This work demonstrates that the SRM neuron not only emulates the adaptive behaviors of a biological neuron,but also enriches the functionality and unleashes the computational power of SNNs.

Force measurement using strain-gauge balance in shock tunnel based on deep learning

When a force test is conducted in a shock tunnel,vibration of the Force Measurement System(FMS)is excited under the strong flow impact,and it cannot be attenuated rapidly within the extremely short test duration of milliseconds order.The output signal of the force balance is coupled with the aerodynamic force and the inertial vibration.This interference can result in inaccurate force measurements,which can negatively impact the accuracy of the test results.To eliminate inertial vibration interference from the output signal,proposed here is a dynamic calibration modeling method for an FMS based on deep learning.The signal is processed using an intelligent Recurrent Neural Network(RNN)model in the time domain and an intelligent Convolutional Neural Network(CNN)model in the frequency domain.Results processed with the intelligent models show that the inertial vibration characteristics of the FMS can be identified efficiently and its main frequency is about 380 Hz.After processed by the intelligent models,the inertial vibration is mostly eliminated from the output signal.Also,the data processing results are subjected to error analysis.The relative error of each component is about 1%,which verifies that the modeling method based on deep learning has considerable engineering application value in data processing for pulse-type strain-gauge balances.Overall,the proposed dynamic calibration modeling method has the potential to improve the accuracy and reliability of force measurements in shock tunnel tests,which could have significant implications for the field of aerospace engineering.

An identifier-actor-optimizer policy learning architecture for optimal control of continuous-time nonlinear systems

An intelligent solution method is proposed to achieve real-time optimal control for continuous-time nonlinear systems using a novel identifier-actor-optimizer(IAO)policy learning architecture.In this IAO-based policy learning approach,a dynamical identifier is developed to approximate the unknown part of system dynamics using deep neural networks(DNNs).Then,an indirect-method-based optimizer is proposed to generate high-quality optimal actions for system control considering both the constraints and performance index.Furthermore,a DNN-based actor is developed to approximate the obtained optimal actions and return good initial guesses to the optimizer.In this way,the traditional optimal control methods and state-of-the-art DNN techniques are combined in the IAO-based optimal policy learning method.Compared to the reinforcement learning algorithms with actor-critic architectures that suffer hard reward design and low computational efficiency,the IAO-based optimal policy learning algorithm enjoys fewer user-defined parameters,higher learning speeds,and steadier convergence properties in solving complex continuous-time optimal control problems(OCPs).Simulation results of three space flight control missions are given to substantiate the effectiveness of this IAO-based policy learning strategy and to illustrate the performance of the developed DNN-based optimal control method for continuous-time OCPs.