Search

A robust deep learning model consisting of long short-term memory and fully connected neural networks has been proposed to automatically interpret homogeneous petroleum reservoirs having infinite, no flow, and constant pressure outer boundary conditions. The pressure change data recorded during the well test operation along with its derivative is input into the model to perform the classification for identifying the reservoir model and, further, regression to estimate output parameter. Gaussian noise was added to analytical models while generating the synthetic training data. The hyperparameters were regulated to perform model optimization, resulting in a batch size of 64, Adam optimization algorithm, learning rate of 0.01, and 80:10:10 data split ratio as the best choices of hyperparameters. The performance accuracy also increased with an increase in the number of samples during training. Suitable classification and regression metrics have been used to evaluate the performance of the models. The paper also demonstrates the prediction performance of the optimized model using simulated and actual oil well pressure drawdown test cases. The proposed model achieved minimum and maximum relative errors of 0.0019 and 0.0308, respectively, in estimating output for the simulated test cases and relative error of 0.0319 for the real test case.

Background We investigated the feasibility of in vitro radiosensitivity prediction with gene expression using deep learning. Methods A microarray gene expression of the National Cancer Institute-60 (NCI-60) panel was acquired from the Gene Expression Omnibus. The clonogenic surviving fractions at an absorbed dose of 2 Gy (SF2) from previous publications were used to measure in vitro radiosensitivity. The radiosensitivity prediction model was based on the convolutional neural network. The 6-fold cross-validation (CV) was applied to train and validate the model. Then, the leave-one-out cross-validation (LOOCV) was applied by using the large-errored samples as a validation set, to determine whether the error was from the high bias of the folded CV. The criteria for correct prediction were defined as an absolute error Results Of the 174 triplicated samples of NCI-60, 171 samples were correctly predicted with the folded CV. Through an additional LOOCV, one more sample was correctly predicted, representing a prediction accuracy of 98.85% (172 out of 174 samples). The average relative error and absolute errors of 172 correctly predicted samples were 1.351±1.875% and 0.00596±0.00638, respectively. Conclusion We demonstrated the feasibility of a deep learning-based in vitro radiosensitivity prediction using gene expression.

In the current work we have used the deep learning based long short-term memory model to predict the strength and peak time of solar cycle 25 by employing the monthly smoothed sunspot number data obtained from WDC-SILSO, Royal Observatory of Belgium, Brussels. We have used the stacked LSTM forecasting model to predict the upcoming cycle 25. From our analysis it has been shown that our proposed model is capable of capturing long term dependencies as well as trend within the data. For cycle 20 and 21 the error difference between predicted as well as observed peak value is 2.3 and 0.7 respectively while the peak prediction error is 1.47% and 0.30%. The RMSE of the model for cycle 20 and 21 is 3.97 and 4.34 respectively. For cycle 22, the AE and RE is 4.6 and 2.16% while the RMSE of the model for this case is 4.50. The predicted peak amplitude of solar cycle 23 and 24 from our proposed model has a relative error of 1.75% and 1.99% respectively from the observed value while the RMSE is 3.4 for cycle 23 and 4.2 for cycle 24. Our projected prediction of cycle 25 using the proposed LSTM model, says that it will be stronger than cycle 24 and weaker than cycle 23. The solar cycle 25 will peak with an amplitude of sunspot number at 171.9 ± 3.4 and will be 47 % stronger than cycle 24. The solar cycle 25 will reach its peak in August 2023 ± 2 months.

This work presents a technique for dielectric property retrieval through Debye parameter reconstruction from open-ended coaxial probe (OECP) response. Debye parameters were obtained with the application of a deep learning (DL) model to the reflection coefficient response of the OECP when terminated with a material under test. The OECP was modelled with the well-known admittance technique from 0.5 to 6 GHz with 20 MHz resolution. A dataset was generated using the admittance technique and obtained data was utilized to design the DL model. As part of the standard procedure, the dataset was separated to train, validate, and test parts by allocating the 80%, 10%, and 10% of the dataset to each section, respectively. Obtained percent relative error for Debye parameters were 1.86±3.01%, 3.33±9.52%, and 2.07±7.42% for $\epsilon _{s}$ , $\epsilon _\infty $ and $\tau $ , respectively. To further test the constructed DL model, OECP responses were measured at the same frequency band when it was terminated with five different standard liquids, four mixtures, and a gel-like material. Reconstructed Debye parameters from the DL model were used to retrieve the complex dielectric properties and obtained results were compared with the literature data. Obtained mean percent relative error was ranging from 1.21±0.06 to 10.89±0.08 within the frequency band of interest.

The bubble point pressure (Pb) is a crucial pressure–volume–temperature (PVT) property and a primary input needed for performing many petroleum engineering calculations, such as reservoir simulation. The industrial practice of determining Pb is by direct measurement from PVT tests or prediction using empirical correlations. The main problems encountered with the published empirical correlations are their lack of accuracy and the noncomprehensive data set used to develop the model. In addition, most of the published correlations have not proven the relationships between the inputs and outputs as part of the validation process (i.e., no trend analysis was conducted). Nowadays, deep learning techniques such as long short-term memory (LSTM) networks have begun to replace the empirical correlations as they generate high accuracy. This study, therefore, presents a robust LSTM-based model for predicting Pb using a global data set of 760 collected data points from different fields worldwide to build the model. The developed model was then validated by applying trend analysis to ensure that the model follows the correct relationships between the inputs and outputs and performing statistical analysis after comparing the most published correlations. The robustness and accuracy of the model have been verified by performing various statistical analyses and using additional data that was not part of the data set used to develop the model. The trend analysis results have proven that the proposed LSTM-based model follows the correct relationships, indicating the model’s reliability. Furthermore, the statistical analysis results have shown that the lowest average absolute percent relative error (AAPRE) is 8.422% and the highest correlation coefficient is 0.99. These values are much better than those given by the most accurate models in the literature.

Non-Intrusive Load Monitoring (NILM) or load disaggregation aims to analyze power consumption by decomposing the energy measured at the aggregate level into constituent appliances level. The conventional load disaggregation framework consists of signal processing and machine learning-based pipelined architectures, respectively for explicit feature extraction and decision making. Manual feature selection in such load disaggregation frameworks leads to biased decisions that eventually reduce system performance. This paper presents an efficient End-to-End (E2E) approach-based unified architecture using Gated Recurrent Units (GRU) for NILM. The proposed approach eliminates explicit feature engineering and has a unified classification and prediction model for appliance power. This eventually reduces the computational cost and enhances response time. The performance of the proposed system is compared with conventional algorithms' with the use of recall, precision, accuracy, F1 score, the relative error in total energy and Mean Absolute Error (MAE). These evaluation metrics are calculated on the power consumption of top priority appliances of Reference Energy Disaggregation Dataset (REDD). The proposed architecture with an overall accuracy of 91.2 and MAE of 25.23 outperforms conventional methods for all electrical appliances. It has been showcased through a series of experiments that feature extraction and event-based approaches for NILM can readily be replaced with E2E deep learning techniques allowing simpler and cost-efficient implementation pathways.

Recently, tremendous progress has been achieved in applying deep learning schemes (DLSs) to solve inverse scattering problems (ISPs), where state-of-the-art performance has been attained. However, little attention has been paid to the uncertainty quantification of these deep-learning methods in solving ISPs. In other words, the error of the prediction is not known since the ground truth is not available in practice. In this article, a Bayesian convolutional neural network (BCNN) is used to quantify the uncertainties in solving ISPs. With Monte Carlo dropout, the proposed BCNN is able to directly predict the pixel-based uncertainties of the widely used DLSs in ISPs. To quantitatively evaluate the performance of uncertainty predictions, both correlation coefficient and nonlinear correlation distribution between the predicted uncertainty and true absolute error are calculated. The tests on both synthetic and experimental data show that the predictive uncertainty is highly correlated with true absolute error calculated from the ground truth. Besides, when DLSs fail to solve an ISP, the predicted uncertainty increases significantly, which offers a pixel-based “confidence level” in solving ISPs.

This paper develops fundamental limits of deep neural network learning by characterizing what is possible if no constraints are imposed on the learning algorithm and on the amount of training data. Concretely, we consider Kolmogorov-optimal approximation through deep neural networks with the guiding theme being a relation between the complexity of the function (class) to be approximated and the complexity of the approximating network in terms of connectivity and memory requirements for storing the network topology and the associated quantized weights. The theory we develop establishes that deep networks are Kolmogorov-optimal approximants for markedly different function classes, such as unit balls in Besov spaces and modulation spaces. In addition, deep networks provide exponential approximation accuracy—i.e., the approximation error decays exponentially in the number of nonzero weights in the network—of the multiplication operation, polynomials, sinusoidal functions, and certain smooth functions. Moreover, this holds true even for one-dimensional oscillatory textures and the Weierstrass function—a fractal function, neither of which has previously known methods achieving exponential approximation accuracy. We also show that in the approximation of sufficiently smooth functions finite-width deep networks require strictly smaller connectivity than finite-depth wide networks., IEEE Transactions on Information Theory, 67 (5), ISSN:0018-9448, ISSN:1557-9654

Advanced and accurate forecasting of COVID‐19 cases plays a crucial role in planning and supplying resources effectively. Artificial Intelligence (AI) techniques have proved their capability in time series forecasting non‐linear problems. In the present study, the relationship between weather factor and COVID‐19 cases was assessed, and also developed a forecasting model using long short‐term memory (LSTM), a deep learning model. The study found that the specific humidity has a strong positive correlation, whereas there is a negative correlation with maximum temperature, and a positive correlation with minimum temperature was observed in various geographic locations of India. The weather data and COVID‐19 confirmed case data (1 April to 30 June 2020) were used to optimize univariate and multivariate LSTM time series forecast models. The optimized models were utilized to forecast the daily COVID‐19 cases for the period 1 July 2020 to 31 July 2020 with 1 to 14 days of lead time. The results showed that the univariate LSTM model was reasonably good for the short‐term (1 day lead) forecast of COVID‐19 cases (relative error

In this paper, we propose a novel deep learning method based on a Convolutional Neural Network (CNN) that simultaneously detects and geolocates plantation-rows while counting its plants considering highly-dense plantation configurations. The experimental setup was evaluated in a cornfield with different growth stages and in a Citrus orchard. Both datasets characterize different plant density scenarios, locations, types of crops, sensors, and dates. A two-branch architecture was implemented in our CNN method, where the information obtained within the plantation-row is updated into the plant detection branch and retro-feed to the row branch; which are then refined by a Multi-Stage Refinement method. In the corn plantation datasets (with both growth phases, young and mature), our approach returned a mean absolute error (MAE) of 6.224 plants per image patch, a mean relative error (MRE) of 0.1038, precision and recall values of 0.856, and 0.905, respectively, and an F-measure equal to 0.876. These results were superior to the results from other deep networks (HRNet, Faster R-CNN, and RetinaNet) evaluated with the same task and dataset. For the plantation-row detection, our approach returned precision, recall, and F-measure scores of 0.913, 0.941, and 0.925, respectively. To test the robustness of our model with a different type of agriculture, we performed the same task in the citrus orchard dataset. It returned an MAE equal to 1.409 citrus-trees per patch, MRE of 0.0615, precision of 0.922, recall of 0.911, and F-measure of 0.965. For citrus plantation-row detection, our approach resulted in precision, recall, and F-measure scores equal to 0.965, 0.970, and 0.964, respectively. The proposed method achieved state-of-the-art performance for counting and geolocating plants and plant-rows in UAV images from different types of crops., Comment: 27 pages, 12 figures, 9 tables

Classical decline methods, such as Arps yield decline curve analysis, have advantages of simple principles and convenient applications, and they are widely used for yield decline analysis. However, for carbonate reservoirs with high initial production, rapid decline, and large production fluctuations, with most wells having no stable production period, the adaptability of traditional decline methods is inadequate. Hence, there is an urgent need to develop a new decline analysis method. Although machine learning methods based on multiple regression and deep learning have been applied to unconventional oil reservoirs in recent years, their application effects have been unsatisfactory. For example, prediction errors based on multiple regression machine learning methods are relatively large, and deep learning sample requirements and the actual conditions of reservoir management do not match. In this study, a new equal probability gene expression programming (EP-GEP) method was developed to overcome the shortcomings of the conventional Arps decline model in the production decline analysis of carbonate reservoirs. Through model validation and comparative analysis of prediction effects, it was proven that the EP-GEP model exhibited good prediction accuracy, and the average relative error was significantly smaller than those of the traditional Arps model and existing machine learning methods. The successful application of the proposed method in the production decline analysis of carbonate reservoirs is expected to provide a new decline analysis tool for field reservoir engineers.

The research of the bulk tobacco flue-curing process is helpful to model the intelligent bulk curing system, which is designed to implement the curing procedure without manual operation. An intelligent bulk curing method based on the convolutional neural network (CNN) named TobaccoNet was proposed, which could set the target dry-bulb temperature ( $T_{\mathrm {D}}$ ) and the target wet-bulb temperature ( $T_{\mathrm {W}}$ ) of the bulk curing barn according to the tobacco leaves image. The performance of the TobaccoNet is compared with the traditional manual image feature extraction method, the stacked-sparse-autoencoder (SSAE)-based deep learning method, and the other two methods applied in related references. The test results show that TobaccoNet outperforms the comparison methods in predicting $T_{\mathrm {D}}$ and $T_{\mathrm {W}}$ . Specifically, the correlation coefficient reaches 0.9965 and 0.9683, the mean relative error is 1.62% and 1.77%, and the root mean squared error achieves 1.061 °C and 0.8581 °C respectively. The promising results demonstrate that TobaccoNet is effective and reliable for modeling the intelligent bulk tobacco flue-curing process. The influence of different CNN structures on the prediction accuracy of $T_{\mathrm {D}}$ and $T_{\mathrm {W}}$ was analyzed. From the perspective of the computational complexity and the prediction performance, the proposed sequential CNN structure is more suitable for analyzing bulk tobacco curing in this study.

As the core power for the aviation industry, shipbuilding industry, and power station industry, it is essential to ensure that the gas turbines operate safely, reliably, greenly and efficiently. Learn from the advantages and disadvantages of the thermodynamic model based and data-driven artificial intelligence based gas-path diagnosis methods, a newfangled gas turbine gas-path diagnosis approach on the basis of knowledge data-driven artificial intelligence is proposed. That is a hybrid method of deep learning and gas path analysis. First, gas turbine thermodynamic model of the object to be diagnosed is constructed by adaptation modeling strategy. And the engine thermodynamic model is taken as the basal model to simulate various gas path faults. Secondly, a large number of knowledge data corresponding to component health parameters and gas turbine boundary condition parameters & gas-path measurable parameters are simulated by setting different component health parameter values and different boundary conditions based on this basal model. And next, define the vector composed of the boundary condition parameters & the gas path measurable parameters in the knowledge database as the input vector, and the component health parameter vector as the output vector, and a deep learning model for regression modeling of this knowledge database is designed. At last, along with the gas turbine engine runs, the trained model outputs component health parameters in real time after trained deep learning model is deployed to the corresponding gas turbine power plant. The simulation experiment results show that, accurate and quantified health parameters of each gas path component can be obtained by the proposed method in this paper, and the overall root mean square error does not exceed 0.033%, and the maximum relative error does not exceed 0.36%, which illustrates the proposed method has great application potential.

The hierarchical deep-learning neural network (HiDeNN) is systematically developed through the construction of structured deep neural networks (DNNs) in a hierarchical manner, and a special case of HiDeNN for representing Finite Element Method (or HiDeNN-FEM in short) is established. In HiDeNN-FEM, weights and biases are functions of the nodal positions, hence the training process in HiDeNN-FEM includes the optimization of the nodal coordinates. This is the spirit of r-adaptivity, and it increases both the local and global accuracy of the interpolants. By fixing the number of hidden layers and increasing the number of neurons by training the DNNs, rh-adaptivity can be achieved, which leads to further improvement of the accuracy for the solutions. The generalization of rational functions is achieved by the development of three fundamental building blocks of constructing deep hierarchical neural networks. The three building blocks are linear functions, multiplication, and inversion. With these building blocks, the class of deep learning interpolation functions are demonstrated for interpolation theories such as Lagrange polynomials, NURBS, isogeometric, reproducing kernel particle method, and others. In HiDeNN-FEM, enrichment functions through the multiplication of neurons is equivalent to the enrichment in standard finite element methods, that is, generalized, extended, and partition of unity finite element methods. Numerical examples performed by HiDeNN-FEM exhibit reduced approximation error compared with the standard FEM. Finally, an outlook for the generalized HiDeNN to high-order continuity for multiple dimensions and topology optimizations are illustrated through the hierarchy of the proposed DNNs.

Robust modeling of a multimodal dynamic system is a challenging and fast-growing area of research. In this study, an integrated bi-modal computing paradigm based on Nonlinear Autoregressive Radial Basis Functions (NAR-RBFs) neural network model, a new family of deep learning with the strength of hybrid artificial neural network, is presented for the solution of nonlinear chaotic dusty system (NCDS) of tiny ionized gas particles arising in fusion devices, industry, astronomy, and space. In the proposed methodology, special transformations are introduced for a class of differential equations, which convert the local optimum to a global optimum. The proposed NAR-RBFs neural network model is implemented on bi-model NCDS represented with Van der Pol-Methiew Equation (VdP-ME) for different scenarios based on variation in dust gain production and loss for both small and large time domains. Excellent agreement for proposed bimodal computing paradigm by the result with the standard state of the arts numerical solvers is verified by attaining RMSE up to 1E−38 for the nonlinear VDP-ME. Accuracy of the proposed model in the critical time domain is also validated by convergence, stability and consistency analysis on statistics calculated from absolute error, root-mean-square error, and analysis of variance metrics.

Among advanced manufacturing techniques for fiber-reinforced polymer-matrix composites (FRPCs) which are critical for aerospace, marine, automotive, and energy industries, frontal polymerization (FP) has been recently proposed to save orders of magnitude of time and energy. However, the cure kinetics of the matrix phase, usually a thermosetting polymer, brings difficulty to the design and control of the process. Here, we develop a deep learning model, ChemNet, to solve an inverse problem for predicting and optimizing the cure kinetics parameters of the thermosetting FRPCs for a desired fabrication strategy. ChemNet consists of a fully connected FeedForward 9 layer deep neural network trained on one million examples, and predicts activation energy and reaction enthalpy given the front characteristics such as speed and maximum temperature. ChemNet provides highly accurate predictions measured by the mean squared error (MSE) and by the maximum absolute error (MAE) metrics. ChemNet's performance on the "hidden" test data set had an MSE of 5.58 × 10-6 and a MAE of 1 × 10-3.

The task of single-channel speech enhancement is to restore clean speech from noisy speech. Recently, speech enhancement has been greatly improved with the introduction of deep learning. Previous work proved that using ideal ratio mask or phase-sensitive mask as intermediation to recover clean speech can yield better performance. In this case, the mean square error is usually selected as the loss function. However, after conducting experiments, the authors find that the mean square error has a problem. It considers absolute error values, meaning that the gradients of the network depend on absolute differences between estimated values and true values, so the points in magnitude spectra with smaller values contribute little to the gradients. To solve this problem, they propose relative loss, which pays more attention to relative differences between magnitude spectra, rather than the absolute differences, and is more in accordance with human sensory characteristics. The perceptual evaluation of speech quality, the short-time objective intelligibility, the signal-to-distortion ratio, and the segmental signal-to-noise ratio are used to evaluate the performance of the relative loss. Experimental results show that it can greatly improve speech enhancement by focusing on smaller values.

Predicting the depth of a scene from monocular infrared images, which plays a crucial role in understanding three-dimensional structures, is one of the challenging tasks in machine learning and computer vision. Considering the lack of texture and color information in infrared images, a novel discrete convolutional conditional random field network is proposed for depth estimation. The proposed method inherits several merits of conditional random fields and deep learning. First, the pairwise features are automatically extracted and optimized through deep architectures. Second, the monocular-images-based depth regression is converted into a multi-class classification, in which the order information of different levels of depths is considered in the loss function. Our experiments demonstrate that this conversion achieves much higher accuracy and faster conversion. Third, to obtain fine-grained level details, we have further proposed a multi-scale discrete convolutional conditional random field network that computes the pairwise features of the discrete conditional random field at different spatial levels. Extensive experiments on the infrared image dataset NUSTMS demonstrate that the proposed method outperforms other depth estimation methods. Specifically, for the proposed method, the mean relative error is 0.181, the mean log10 error is 0.072, and the accuracy with a threshold (t = 1.253) is 95.3%.

In this study, we consider the application of deep learning (DL) to tabu search (TS) detection in large multiple-input multiple-output (MIMO) systems. First, we propose a deep neural network (DNN) architecture for symbol detection, termed the fast-convergence sparsely connected detection network (FS-Net), which is obtained by optimizing the prior detection networks called DetNet and ScNet. Then, we propose the DL-aided TS algorithm, in which the initial solution is approximated by the proposed FS-Net. Furthermore, in this algorithm, an adaptive early termination (ET) algorithm and a modified searching process are performed based on the predicted approximation error, which is determined from the FS-Net-based initial solution, so that the optimal solution can be reached earlier. The simulation results show that the proposed algorithm achieves approximately 90% complexity reduction for a $32 \times 32$ MIMO system with QPSK with respect to the existing TS algorithms, while maintaining almost the same performance.

Introduction: The purpose of this work was to assess the feasibility of acquisition time reduction in MPI-SPECT imaging using deep leering techniques through two main approaches, namely reduction of the acquisition time per projection and reduction of the number of angular projections. Methods: SPECT imaging was performed using a fixed 90° angle dedicated dual-head cardiac SPECT camera. This study included a prospective cohort of 363 patients with various clinical indications (normal, ischemia, and infarct) referred for MPI-SPECT. For each patient, 32 projections for 20 seconds per projection were acquired using a step and shoot protocol from the right anterior oblique to the left posterior oblique view. SPECT projection data were reconstructed using the OSEM algorithm (6 iterations, 4 subsets, Butterworth post-reconstruction filter). For each patient, four different datasets were generated, namely full time (20 seconds) projections (FT), half-time (10 seconds) acquisition per projection (HT), 32 full projections (FP), and 16 half projections (HP). The image-to-image transformation via the residual network was implemented to predict FT from HT and predict FP from HP images in the projection domain. Qualitative and quantitative evaluations of the proposed framework was performed using a tenfold cross validation scheme using the root mean square error (RMSE), absolute relative error (ARE), structural similarity index, peak signal-to-noise ratio (PSNR) metrics, and clinical quantitative parameters. Results: The results demonstrated that the predicted FT had better image quality than the predicted FP images. Among the generated images, predicted FT images resulted in the lowest error metrics (RMSE = 6.8 ± 2.7, ARE = 3.1 ± 1.1%) and highest similarity index and signal-to-noise ratio (SSIM = 0.97 ± 1.1, PSNR = 36.0 ± 1.4). The highest error metrics (RMSE = 32.8 ± 12.8, ARE = 16.2 ± 4.9%) and the lowest similarity and signal-to-noise ratio (SSIM = 0.93 ± 2.6, PSNR = 31.7 ± 2.9) were observed for HT images. The RMSE decreased significantly (P value

Obtaining high-quality phenotypic data that can be used to study the relationship between genotype, phenotype and environment is still labour-intensive. Digital plant phenotyping can assist in collecting these data by replacing human vision by computer vision. However, for complex traits, such as plant architecture, robust and generic digital phenotyping methods have not yet been developed. This study focusses on internode length in cucumber plants. A method for estimating internode length and internode development over time is proposed. The proposed method firstly applies a robust node-detection algorithm based on a deep convolutional neural network. In tests, the algorithm had a precision of 0.95 and a recall of 0.92. The nodes are detected in images from multiple viewpoints around the plant in order to deal with the complex and cluttered plant environment and to solve the occlusion of nodes by other plant parts. The nodes detected in the multiple viewpoint images are then clustered using affinity propagation. The predicted clusters had a homogeneity of 0.98 and a completeness of 0.99. Finally, a linear function is fitted, which allows to study internode development over time. The presented method was able to measure internode length in cucumber plants with a higher accuracy and a larger temporal resolution than other methods proposed in literature and without the time investment needed to obtain the measurements manually. The relative error of our complete method was 5.8%. The proposed method provides many opportunities for robust phenotyping of fruit-vegetable crops grown under greenhouse conditions.

Accurately modeling and predicting wireless channelquality variations is essential for a number of networking applications such as scheduling and improved video streaming over 4G LTE networks and bit rate adaptation for improved performance in WiFi networks. In this paper, we design DeepChannel, an encoder-decoder based sequence-to-sequence deep learning model that is capable of predicting future wireless signal strength variations based on past signal strength data. We consider two different versions of DeepChannel; the first and second versions use LSTM and GRU as their basic cell structure, respectively. In contrast to prior work that is primarily focused on designing models for particular network settings, DeepChannel is highly adaptable and can predict future channel conditions for different networks, sampling rates, mobility patterns, and communication standards. We compare the performance (i.e., the root mean squared error, mean absolute error and relative error of future predictions) of DeepChannel with respect to two baselines— i) linear regression, and ii) ARIMA for multiple networks and communication standards. In particular, we consider 4G LTE, WiFi, WiMAX, an industrial network operating in the 5.8 GHz range, and Zigbee networks operating under varying levels of user mobility and observe that DeepChannel provides significantly superior performance. Finally, we provide a detailed discussion of the key design decisions including insights into hyper-parameter tuning and the applicability of our model in other networking scenarios.

Influenced by climate change and urbanization, urban flood frequently occurs and represents a serious challenge for many cities. Therefore, it is necessary to generate refined predictions of urban floods, such as the prediction of water accumulation processes at water accumulation points, which is of great significance for supporting water-related managers to reduce flood losses. In this study, 16 combination schemes of rainfall sensitivity indicators were used to determine the optimal scheme for predicting the depth of accumulated water, and the gradient boosting decision tree (GBDT) algorithm in deep learning was used to build a prediction model of the accumulation process of urban stormy accumulation points. Among the 16 schemes, the relative error of scheme 1 is 15.39%, and the qualified rate is 92.86%. This scheme exhibits the highest accuracy for the prediction results of water accumulation depth. Given this finding, the GBDT algorithm was used to construct a regression prediction model of the water accumulation process based on the collected historical rainfall water accumulation data of 50 water accumulation points. The results demonstrated that the GBDT regression prediction model has a mean relative error of 19.77%, a qualified rate of 82.00%, and a peak average relative error of 5.48%, which verify the validity and applicability of the model for the real-time prediction of the process of water accumulation.

Purpose Better forecasting always leads to better management and planning of the operations. The container throughput data are complex and often have multiple seasonality. This makes it difficult to forecast accurately. The purpose of this paper is to forecast container throughput using deep learning methods and benchmark its performance over other traditional time-series methods. Design/methodology/approach In this study, long short-term memory (LSTM) networks are implemented to forecast container throughput. The container throughput data of the Port of Singapore are used for empirical analysis. The forecasting performance of the LSTM model is compared with seven different time-series forecasting methods, namely, autoregressive integrated moving average (ARIMA), simple exponential smoothing, Holt–Winter’s, error-trend-seasonality, trigonometric regressors (TBATS), neural network (NN) and ARIMA + NN. The relative error matrix is used to analyze the performance of the different models with respect to bias, accuracy and uncertainty. Findings The results showed that LSTM outperformed all other benchmark methods. From a statistical perspective, the Diebold–Mariano test is also conducted to further substantiate better forecasting performance of LSTM over other counterpart methods. Originality/value The proposed study is a contribution to the literature on the container throughput forecasting and adds value to the supply chain theory of forecasting. Second, this study explained the architecture of the deep-learning-based LSTM method and discussed in detail the steps to implement it.

Among the several paradigms of artificial intelligence (AI) or machine learning (ML), a remarkably successful paradigm is deep learning. Deep learning's phenomenal success has been hoped to be interpreted via fundamental research on the theory of deep learning. Accordingly, applied research on deep learning has spurred the theory of deep learning-oriented depth and breadth of developments. Inspired by such developments, we pose these fundamental questions: can we accurately approximate an arbitrary matrix-vector product using deep rectified linear unit (ReLU) feedforward neural networks (FNNs)? If so, can we bound the resulting approximation error? In light of these questions, we derive error bounds in Lebesgue and Sobolev norms that comprise our developed deep approximation theory. Guided by this theory, we have successfully trained deep ReLU FNNs whose test results justify our developed theory. The developed theory is also applicable for guiding and easing the training of teacher deep ReLU FNNs in view of the emerging teacher-student AI or ML paradigms that are essential for solving several AI or ML problems in wireless communications and signal processing; network science and graph signal processing; and network neuroscience and brain physics.

We derive bounds on the error, in high-order Sobolev norms, incurred in the approximation of Sobolev-regular as well as analytic functions by neural networks with the hyperbolic tangent activation function. These bounds provide explicit estimates on the approximation error with respect to the size of the neural networks. We show that tanh neural networks with only two hidden layers suffice to approximate functions at comparable or better rates than much deeper ReLU neural networks., Neural Networks, 143

Timely and accurate regional crop-yield estimates are crucial for guiding agronomic practices and policies to improve food security. In this study, a crop-growth model was integrated with time series of remotely sensed data through deep learning (DL) methods to improve the accuracy of regional wheat-yield estimations in Henan Province, China. Firstly, the time series of moderate-resolution imaging spectroradiometer (MODIS) normalized difference vegetation index (NDVI) were input into the long short-term memory network (LSTM) model to identify the wheat-growing region, which was further used to estimate wheat areas at the municipal and county levels. Then, the leaf area index (LAI) and grain-yield time series simulated by the Crop Environment REsource Synthesis for Wheat (CERES-Wheat) model were used to train and evaluate the LSTM, one-dimensional convolutional neural network (1-D CNN) and random forest (RF) models, respectively. Finally, an exponential model of the relationship between the field-measured LAI and MODIS NDVI was applied to obtain the regional LAI, which was input into the trained LSTM, 1-D CNN and RF models to estimate wheat yields within the wheat-growing region. The results showed that the linear correlations between the estimated wheat areas and the statistical areas were significant at both the municipal and county levels. The LSTM model provided more accurate estimates of wheat yields, with higher R2 values and lower root mean square error (RMSE) and mean relative error (MRE) values than the 1-D CNN and RF models. The LSTM model has an inherent advantage in capturing phenological information contained in the time series of the MODIS-derived LAI, which is important for satellite-based crop-yield estimates.

Conventional acoustic metasurfaces are constructed with gradiently "local" phase shift profiles provided by subunits. The local strategy implies the ignorance of the mutual coupling between subunits, which limits the efficiency of targeted sound manipulation, especially in complex environments. By taking into account the "nonlocal" interaction among subunits, nonlocal metasurface offers an opportunity for accurate control of sound propagation, but the requirement of the consideration of gathering coupling among all subunits, not just the nearest-neighbor coupling, greatly increases the complexity of the system and therefore hinders the explorations of functionalities of nonlocal metasurfaces. In this work, empowered by deep learning algorithms, the complex gathering coupling can be learned efficiently from the preset dataset so that the functionalities of nonlocal metasurfaces can be significantly uncovered. As an example, we demonstrate that nonlocal metasurfaces, which can redirect an incident wave into multi-channel reflections with arbitrary energy ratios, can be accurately predicted by deep learning algorithms. Compared to the theory, the relative error of the energy ratios is less than 1\%. Furthermore, experiments witness three-channel reflection with three types of energy ratios of (1, 0, 0), (1/2, 0, 1/2), and (1/3, 1/3, 1/3), proving the validity of the deep learning enabled nonlocal metasurfaces. Our work might blaze a new trail in the design of acoustic functional devices, especially for the cases containing complex wave-matter interactions.

Purpose Radiation therapy treatment planning is a trial-and-error, often time-consuming process. An approximately optimal dose distribution corresponding to a specific patient's anatomy can be predicted by using pre-trained deep learning (DL) models. However, dose distributions are often optimized based not only on patient-specific anatomy but also on physicians' preferred trade-offs between planning target volume (PTV) coverage and organ at risk (OAR) sparing or among different OARs. Therefore, it is desirable to allow physicians to fine-tune the dose distribution predicted based on patient anatomy. In this work, we developed a DL model to predict the individualized 3D dose distributions by using not only the patient's anatomy but also the desired PTV/OAR trade-offs, as represented by a dose volume histogram (DVH), as inputs. Methods In this work, we developed a modified U-Net network to predict the 3D dose distribution by using patient PTV/OAR masks and the desired DVH as inputs. The desired DVH, fine-tuned by physicians from the initially predicted DVH, is first projected onto the Pareto surface, then converted into a vector, and then concatenated with feature maps encoded from the PTV/OAR masks. The network output for training is the dose distribution corresponding to the Pareto optimal DVH. The training/validation datasets contain 77 prostate cancer patients, and the testing dataset has 20 patients. Results The trained model can predict a 3D dose distribution that is approximately Pareto optimal while having the DVH closest to the input desired DVH. We calculated the difference between the predicted dose distribution and the optimized dose distribution that has a DVH closest to the desired one for the PTV and for all OARs as a quantitative evaluation. The largest absolute error in mean dose was about 3.6% of the prescription dose, and the largest absolute error in the maximum dose was about 2.0% of the prescription dose. Conclusions In this feasibility study, we have developed a 3D U-Net model with the patient's anatomy and the desired DVH curves as inputs to predict an individualized 3D dose distribution that is approximately Pareto optimal while having the DVH closest to the desired one. The predicted dose distributions can be used as references for dosimetrists and physicians to rapidly develop a clinically acceptable treatment plan.

An advanced method of modeling radio-frequency (RF) devices based on a deep learning technique is proposed for accurate prediction of S parameters. The S parameters of RF devices calculated by full-wave electromagnetic solvers along with the metallic geometry of the structure, permittivity and thickness of the dielectric layers of the RF devices are used partly as the training and partly as testing data for the deep learning structure. To implement the training procedure efficiently, a novel selection method of training data considering critical points is introduced. In order to rapidly and accurately map the geometrical parameters of the RF devices to the S parameters, deep neural networks are used to establish the multiple non-linear transforms. The hidden-layers of the neural networks are adaptively chosen based on the frequency response of the RF devices to guarantee the accuracy of generated model. The Adam optimization algorithm is utilized for the acceleration of training. With the established deep learning model of a parameterized device, the S parameters can efficiently be obtained when the device geometrical parameters change. Comparing with the traditional modeling method that uses shallow neural networks, the proposed method can achieve better accuracy, especially when the training data are non-uniform. Three RF devices, including a rectangular inductor, an interdigital capacitor, and two coupled transmission lines, are used for building and verifying the deep neural network. It is shown that the deep neural network has good robustness and excellent generalization ability. Even for very wide frequency band (0–100 GHz), the maximum relative error of the coupled transmission lines using the proposed method is below 3%.

The aboveground fresh weight of weeds is an important indicator that reflects their biomass and physiological activity and directly affects the criteria for determining the amount of herbicides to apply. In precision agriculture, the development of models that can accurately locate weeds and predict their fresh weight can provide visual support for accurate, variable herbicide application in real time. In this work, we develop a two-stream dense feature fusion convolutional network model based on RGB-D data for the real-time prediction of the fresh weight of weeds. A data collection method is developed for the compilation and production of RGB-D data sets. The acquired images undergo data enhancement, and a depth transformation data enhancement method suitable for depth data is proposed. The main idea behind the approach in this study is to use the YOLO-V4 model to locate weeds and use the two-stream dense feature fusion network to predict their aboveground fresh weight. In the two-stream dense feature fusion network, DenseNet and NiN methods are used to construct a Dense-NiN-Block structure for deep feature extraction and fusion. The Dense-NiN-Block module was embedded in five convolutional neural networks for comparison, and the best results were achieved with DenseNet201. The test results show that the predictive ability of the convolutional network using RGB-D as the input is better than that of the network using RGB as the input without the Dense-NiN-Block module. The mAP of the proposed network is 75.34% (IoU value of 0.5), the IoU is 86.36%, the detection speed of the fastest model with a RTX2080Ti NVIDIA graphics card is 17.8 fps, and the average relative error is approximately 4%. The model proposed in this paper can provide visual technical support for precise, variable herbicide application. The model can also provide a reference method for the non-destructive prediction of crop fresh weight in the field and can contribute to crop breeding and genetic improvement.