NN-based Approach Research Articles

Neural network-based reconstruction of steady-state temperature systems with unknown material composition

Effective thermal management is crucial for the performance and longevity of devices such as computer chips and batteries. A fundamental challenge in this field is accurately determining temperature distributions, which are often limited by incomplete observations. We introduce a neural-network-based approach to overcome these challenges. Specifically, a physics-informed fully convolutional auto-encoder coupled with a preceding spatial propagator network module is used to reconstruct complete temperature fields from partial observations, without requiring additional information about the spatial distribution of materials with varying thermal properties. This paper details the development and validation of our model, demonstrating its effectiveness in virtual sensing setups for both simulated and real-world thermal systems. The reconstruction capabilities for different sampling strategies are investigated. The most challenging of which is the “grid-edge” sampling where only discrete temperature sensors on the periphery of the systems are taken as input, amounting to a sampling ratio as low as 3.12%. The presented method is able to reconstruct the full temperature field with a relative average error of 1.1%. The NN-based approach outperforms the standard Kriging method in predicting the maximum temperature in a system, and proves more robust when the available measurements contain little information.

Artificial neural network for deciphering the structural transformation of condensed ZnO by extended x-ray absorption fine structure spectroscopy

Pressure-induced structural phase transitions play a pivotal role in unlocking novel material functionalities and facilitating innovations in materials science. Nonetheless, unveiling the mechanisms of densification, which relies heavily on precise and comprehensive structural analysis, remains a challenge. Herein, we investigated the archetypal B4 → B1 phase transition pathway in ZnO by combining x-ray absorption fine structure (XAFS) spectroscopy with machine learning. Specifically, we developed an artificial neural network (NN) to decipher the extended-XAFS spectra by reconstructing the partial radial distribution functions of Zn–O/Zn pairs. This provided us with access to the evolution of the structural statistics for all the coordination shells in condensed ZnO, enabling us to accurately track the changes in the internal structural parameter u and the anharmonic effect. We observed a clear decrease in u and an increased anharmonicity near the onset of the B4 → B1 phase transition, indicating a preference for the iT phase as the intermediate state to initiate the phase transition that can arise from the softening of shear phonon modes. This study suggests that NN-based approach can facilitate a more comprehensive and efficient interpretation of XAFS under complex in-situ conditions, which paves the way for highly automated data processing pipelines for high-throughput and real-time characterizations in next-generation synchrotron photon sources.

A Deep Learning-based Model to Predict Limb Amputation in Peripheral Vascular Trauma

The objective of this study is to develop an accurate neural network (NN)-based approach to predict limb amputation in patients who sustain blunt traumatic peripheral vascular injury. The 2019 National Trauma Data Bank (NTDB) was queried for International Classification of Disease, Tenth Edition (ICD-10) codes corresponding to blunt peripheral vascular trauma. A total of 15,438 patients were used to train a NN, of which 335 received a traumatic amputation. The neural network was trained using the following input variables: age, sex, Injury Severity Scale (ISS), presenting Glasgow Coma Scale (GCS), packed red blood cells (RBCs) used in the first 4 hours of presentation, heart rate, and presenting systolic blood pressure (SBP). The primary outcome of our study was the area under the receiver-operator characteristics curve (AUC) of our NN with secondary outcomes including the normalized importance ratio of variables used to train the NN. The NN accurately predicted amputation in patients who sustained blunt peripheral vascular injury with an AUC of 0.78. There were no significant differences in age between the amputation and limb salvage cohorts. There was a higher prevalence of males in the amputation cohort in addition to a higher ISS, lower GCS, increased amount of packed RBCs in the first 4 hours of presentation, and a lower presenting SBP. Our NN found ISS followed by blood transfusion requirements to be most predictive of traumatic amputation. A NN-based approach can accurately predict. Additionally, presenting ISS and transfusion requirements within the first 4 hours were found to have the highest predictive value. Furthermore, NN can be utilized as an innovative predictive adjunct in the management of critical trauma.

Predicting Mortality in Blunt Abdominal Aortic Trauma- a Novel Approach Utilizing Neural Networks

The aim of this study was to develop an accurate neural network (NN)-based approach to predict in-hospital mortality in patients who sustain blunt traumatic aortic injury (BTAI). Secondary objectives include identifying factors that may play a contributory role in mortality. The 2019 National Trauma Data Bank (NTDB) was queried for International Classification of Diseases, Tenth Edition (ICD-10) codes corresponding to blunt abdominal aortic trauma. Patients with no signs of life upon presentation were excluded from this study. A total of 113 patients were used to train a NN using the following input variables: age, sex, Injury Severity Scale (ISS), presenting Glasgow Coma Scale (GCS), packed red blood cells (RBCs) used in the first 4 hours of presentation, and presenting systolic blood pressure (SBP). The primary outcome of our study was the area under the receiver-operator characteristics curve (AUC), sensitivity, and specificity of our NN with secondary outcomes including the normalized importance ratio of variables used to train the NN. The NN accurately predicted mortality in patients who sustained BTAI with an AUC of 0.93 (Fig), sensitivity of 72%, and specificity of 89%. There were no significant differences in age between the mortality and survival cohorts. There was a higher prevalence of males in the mortality group in addition to a higher ISS, lower GCS, increased number of packed RBCs in the first 4 hours of presentation and a lower presenting SBP. Our NN found transfusion requirements to be the most predictive of mortality followed by ISS. A NN-based approach can accurately predict mortality in BTAI. Additionally, transfusions requirements within the first 4 hours and presenting ISS were found to have the highest predictive value. Furthermore, NN can be utilized as an innovative predictive adjunct in the management of critical trauma.

Interpretable Machine Learning on Metabolomics Data Reveals Biomarkers for Parkinson's Disease.

The use of machine learning (ML) with metabolomics provides opportunities for the early diagnosis of disease. However, the accuracy of ML and extent of information obtained from metabolomics can be limited owing to challenges associated with interpreting disease prediction models and analyzing many chemical features with abundances that are correlated and "noisy". Here, we report an interpretable neural network (NN) framework to accurately predict disease and identify significant biomarkers using whole metabolomics data sets without a priori feature selection. The performance of the NN approach for predicting Parkinson's disease (PD) from blood plasma metabolomics data is significantly higher than other ML methods with a mean area under the curve of >0.995. PD-specific markers that predate clinical PD diagnosis and contribute significantly to early disease prediction were identified including an exogenous polyfluoroalkyl substance. It is anticipated that this accurate and interpretable NN-based approach can improve diagnostic performance for many diseases using metabolomics and other untargeted 'omics methods.

A New NN-Based Approach to In-Sensor PDM-to-PCM Conversion for Ultra TinyML KWS

This brief proposes a new approach based on a tiny neural network to convert Pulse Density Modulation (PDM) signals acquired from digital Micro-Electro-Mechanical System (MEMS) microphones into the standard Pulse Code Modulation (PCM) format for any further digital audio processing. The proposed approach allows for a compact and ultra-low-power hardware implementation of the conversion, suitable for ultra tinyML Key Word Spotting (KWS) applications, closely coupled with the sensor itself and tightly coupled with a neural network classifier. The converter achieves a signal-to-noise ratio value of 48 dB, which enables KWS accuracy of 89% over 12 classes. Implementation on a Xilinx Artix-7 FPGA results in 917 LUTs, 361 FFs, and 182 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu \text{W}$ </tex-math></inline-formula> Dynamic Power (DynP) consumption. By targeting the TSMC 0.13 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu \text{m}$ </tex-math></inline-formula> CMOS technology, the synthesis reports an area occupation of 0.086 mm2 and a DynP of 128.7 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu \text{W}$ </tex-math></inline-formula> /MHz. These results enable the integration of the proposed design into the CMOS circuitry closely coupled with the MEMS microphone.

FASCINET: A Fully Automated Single-Board Computer Generator Using Neural Networks

Designing single-board computers (SBCs) is becoming more challenging given the growing number of discrete components that are made available and the rate at which this number grows. Keeping track of all available components options, revisions, and functionalities is challenging for SBC designers who are striving for faster design cycles. Moreover, the procedure of deciding peripheral components, their values, and connections of an SBC is not only difficult because of various parameters that need to be considered but also is time consuming as there exist numerous components on a typical SBC nowadays. In this article, an SBC generator tool, FASCINET, is presented that uses a neural network (NN) model to design customized peripheral circuits for SBCs. The tool creates a large commercial off-the-shelf database (COTS DB) of existing components, efficiently searches through them, and selects optimal components for both main and peripheral components based on the user’s requirements. Creating such a broad COTS DB requires processing abundant datasheets. A manual approach is time consuming, even if only a fraction of all available datasheets is considered. In order to automate this process, this article describes a novel NN-based approach for automatically categorizing datasheets and proposes an extraction technique for parsing relevant functional information from tables within. Our evaluation using a test set that contains over 770 000 components shows that the category of datasheets is identified correctly over 95% of the time. Additionally, the table extractor has a precision above 96%. Our proposed fully autonomous SBC design approach reduces the time for generating the schematic of an SBC to as little as 2 min. For validating the accuracy of our model, the netlists of 400 SBCs designed by FASCINET are compared to the human-designed versions. This evaluation shows that FASCINET is able to design SBCs that are identical to the manually designed ones except for minor differences.

A low-cost virtual coach for 2D video-based compensation assessment of upper extremity rehabilitation exercises

BackgroundThe increasing demands concerning stroke rehabilitation and in-home exercise promotion grew the need for affordable and accessible assistive systems to promote patients’ compliance in therapy. These assistive systems require quantitative methods to assess patients’ quality of movement and provide feedback on their performance. However, state-of-the-art quantitative assessment approaches require expensive motion-capture devices, which might be a barrier to the development of low-cost systems.MethodsIn this work, we develop a low-cost virtual coach (VC) that requires only a laptop with a webcam to monitor three upper extremity rehabilitation exercises and provide real-time visual and audio feedback on compensatory motion patterns exclusively from image 2D positional data analysis. To assess compensation patterns quantitatively, we propose a Rule-based (RB) and a Neural Network (NN) based approaches. Using the dataset of 15 post-stroke patients, we evaluated these methods with Leave-One-Subject-Out (LOSO) and Leave-One-Exercise-Out (LOEO) cross-validation and the F_1 score that measures the accuracy (geometric mean of precision and recall) of a model to assess compensation motions. In addition, we conducted a pilot study with seven volunteers to evaluate system performance and usability.ResultsFor exercise 1, the RB approach assessed four compensation patterns with a F_1 score of 76.69 %. For exercises 2 and 3, the NN-based approach achieved a F_1 score of 72.56 % and 79.87 %, respectively. Concerning the user study, they found that the system is enjoyable (hedonic value of 4.54/5) and relevant (utilitarian value of 4.86/5) for rehabilitation administration. Additionally, volunteers’ enjoyment and interest (Hedonic value perception) were correlated with their perceived VC performance (rho = 0.53).ConclusionsThe VC performs analysis on 2D videos from a built-in webcam of a laptop and accurately identifies compensatory movement patterns to provide corrective feedback. In addition, we discuss some findings concerning system performance and usability.

Nonlinear Dynamics Modeling and Fault Detection for a Soft Trunk Robot: An Adaptive NN-Based Approach

Nonlinear Dynamics Modeling and Fault Detection for a Soft Trunk Robot: An Adaptive NN-Based Approach

Radioactive hot-spot localisation and identification using deep learning

The detection of radioactive hot-spots and the identification of the radionuclides present have been a challenge for the security sector, especially in situations involving chemical, biological, radiological, nuclear and explosive threats, as well as naturally occurring radioactive materials. This work proposes a solution based on Machine Learning techniques, with a focus on artificial neural networks (NNs), in order to localise, quantify and identify radioactive sources. Firstly, the created RHLnet model uses observations of radiological intensity counts and corresponding localisations to estimate the number, location and activity of unknown radioactive sources present in a given scenario. Then, another model (RHIdnet) gets the gamma spectrum of the sources to perform the identification of the corresponding radionuclides. For this, a training data set composed of simulated data is used during the training process, and so, using algorithms with the models already trained, fast and accurate predictions are achieved, ensuring the reliability of such a NN-based approach. The proposed solution is tested in simulated and real scenarios, with multiple sources, providing a low number of limitations, related to possible false negatives and false positives. Besides, the results have shown that the algorithm is scalable for very large regions, as well as for very small scenarios. Single and multiple isotope identification on each sample is explored, highlighting the benefits as well as possible improvements. Thus, NNs have demonstrated the capability of being an emerging tool with the potential to make a difference in the nuclear field, by helping in the development of novel techniques and new solutions in order to safeguard human lives.

A Predictive Analysis of Wall Stress in Abdominal Aortic Aneurysms Using a Neural Network Model.

Rupture risk assessment of abdominal aortic aneurysms (AAAs) by means of quantifying wall stress is a common biomechanical strategy. However, the clinical translation of this approach has been greatly limited due to the complexity associated with the computational tools required for its implementation. Thus, being able to estimate wall stress using nonbiomechanical markers that can be quantified as a direct outcome of clinical image segmentation would be advantageous in improving the potential implementation of said strategy. In the present work, we investigated the use of geometric indices to predict patient-specific AAA wall stress by means of a novel neural network (NN) modeling approach. We conducted a retrospective review of existing clinical images of two patient groups: 98 asymptomatic and 50 symptomatic AAAs. The images were subject to a protocol consisting of image segmentation, processing, volume meshing, finite element modeling, and geometry quantification, from which 53 geometric indices and the spatially averaged wall stress (SAWS) were calculated. SAWS estimated from finite element analysis was considered the gold standard for the predictions. We developed feed-forward NN models composed of an input layer, two dense layers, and an output layer using Keras, a deep learning library in python. The NN models were trained, tested, and validated independently for both AAA groups using all geometric indices, as well as a reduced set of indices resulting from a variable reduction procedure. We compared the performance of the NN models with two standard machine learning algorithms (MARS: multivariate adaptive regression splines and GAM: generalized additive model) and a linear regression model (GLM: generalized linear model). With the reduced sets of indices, the NN-based approach exhibited the highest mean goodness-of-fit (for the symptomatic group 0.71 and for the asymptomatic group 0.79) and lowest mean relative error (17% for both groups). In contrast, MARS yielded a mean goodness-of-fit of 0.59 for the symptomatic group and 0.77 for the asymptomatic group, with relative errors of 17% for the symptomatic group and 22% for the asymptomatic group. GAM had a mean goodness-of-fit of 0.70 for the symptomatic group and 0.80 for the asymptomatic group, with relative errors of 16% for the symptomatic group and 20% for the asymptomatic group. GLM did not perform as well as the other algorithms, with a mean goodness-of-fit of 0.53 for the symptomatic group and 0.70 for the asymptomatic group, with relative errors of 19% for the symptomatic group and 23% for the asymptomatic group. Nevertheless, the NN models required a reduced set of 15 and 13 geometric indices to predict SAWS for the symptomatic and asymptomatic AAA groups, respectively. This was in contrast to the reduced set of nine and eight geometric indices required to predict SAWS with the MARS and GAM algorithms for each AAA group, respectively. The use of NN modeling represents a promising alternative methodology for the estimation of AAA wall stress using geometric indices as surrogates, in lieu of finite element modeling. The performance metrics of NN models are expected to improve with significantly larger group sizes, given the suitability of NN modeling for "big data" applications.

Embedding and classifying test execution traces using neural networks

Classifying test executions automatically as pass or fail remains a key challenge in software testing and is referred to as the test oracle problem. It is being attempted to solve this problem with supervised learning over test execution traces. A programme is instrumented to gather execution traces as sequences of method invocations. A small fraction of the programme's execution traces is labelled with pass or fail verdicts. Execution traces are then embedded as fixed length vectors and a neural network (NN) component that uses the line-by-line information to classify traces as pass or fail is designed. The classification accuracy of this approach is evaluated using subject programs from different application domains—1. Module from Ethereum Blockchain, 2. Module from PyTorch deep learning framework, 3. Microsoft SEAL encryption library components, 4. Sed stream editor, 5. Nine network protocols from Linux packet identifier, L7-Filter and 6. Utilities library, commons-lang for Java. For all subject programs, it was found that test execution classification had high precision, recall and specificity, averaging to 93%, 94% and 96%, respectively, while only training with an average 14% of the total traces. Experiments show that the proposed NN-based approach is promising in classifying test executions from different application domains.

OFDM PAPR Reduction via Time-Domain Scattered Sampling and Hybrid Batch Training of Synchronous Neural Networks

Peak-to-average power ratio (PAPR) reduction in multiplexed signals in orthogonal frequency division multiplexing (OFDM) systems has been a long-standing critical issue. Clipping and filtering (CF) techniques offer good performance in terms of PAPR reduction at the expense of a relatively high computational cost that is inherent in the repeated application of fast Fourier transform (FFT) operations. The ever-increasing demand for low-latency operation calls for the development of low-complexity novel solutions to the PAPR problem. To address this issue while providing an enhanced PAPR reduction performance, we propose a synchronous neural network (NN)-based solution to achieve PAPR reduction performance exceeding the limits of conventional CF schemes with lower computational complexity. The proposed scheme trains a neural network module using hybrid collections of samples from multiple OFDM symbols to arrive at a signal mapping with desirable characteristics. The benchmark NN-based approach provides a comparable performance to conventional CF. However, it can underfit or overfit due to its asynchronous nature which leads to increased out-of-band (OoB) radiations, and deteriorating bit error rate (BER) performance for high-order modulations. Simulations’ results demonstrate the effectiveness of the proposed scheme in terms of the achieved cubic metric (CM), BER, and OoB emissions.

Design of one-dimensional acoustic metamaterials using machine learning and cell concatenation

Metamaterial systems have opened new, unexpected, and exciting paths for the design of acoustic devices that only few years ago were considered completely out of reach. However, the development of an efficient design methodology still remains challenging due to highly intensive search in the design space required by the conventional optimization-based approaches. To address this issue, this study develops two machine learning (ML)-based approaches for the design of one-dimensional periodic and non-periodic metamaterial systems. For periodic metamaterials, a reinforcement learning (RL)-based approach is proposed to design a metamaterial that can achieve user-defined frequency band gaps. This RL-based approach surpasses conventional optimization-based methods in the reduction of computation cost when a near-optimal solution is acceptable. Leveraging the capability of exploration in RL, the proposed approach does not require any training datasets generation and therefore can be deployed for online metamaterial design. For non-periodic metamaterials, a neural network (NN)-based approach capable of learning the behavior of individual material units is presented. By assembling the NN representation of individual material units, a surrogate model of the whole metamaterial is employed to determine the properties of the resulting assembly. Interestingly, the proposed approach is capable of modeling different metamaterial assemblies satisfying user-defined properties while requiring only a one-time network training procedure. Also, the NN-based approach does not need a pre-defined number of material unit cells, and it works when the physical model of the unit cell is not well understood, or the situation where only the sensor measurements of the unit cell are available. The robustness of the proposed two approaches is validated through numerical simulations and design examples.

Exploiting Deep Learning for Secure Transmission in an Underlay Cognitive Radio Network

This paper investigates a machine learning-based power allocation design for secure transmission in a cognitive radio (CR) network. In particular, a neural network (NN)-based approach is proposed to maximize the secrecy rate of the secondary receiver under the constraints of total transmit power of secondary transmitter, and the interference leakage to the primary receiver, within which three different regularization schemes are developed. The key advantage of the proposed algorithm over conventional approaches is the capability to solve the power allocation problem with both perfect and imperfect channel state information. In a conventional setting, two completely different optimization frameworks have to be designed, namely the robust and non-robust designs. Furthermore, conventional algorithms are often based on iterative techniques, and hence, they require a considerable number of iterations, rendering them less suitable in future wireless networks where there are very stringent delay constraints. To meet the unprecedented requirements of future ultra-reliable low-latency networks, we propose an NN-based approach that can determine the power allocation in a CR network with significantly reduced computational time and complexity. As this trained NN only requires a small number of linear operations to yield the required power allocations, the approach can also be extended to different delay sensitive applications and services in future wireless networks. When evaluate the proposed method versus conventional approaches, using a suitable test set, the proposed approach can achieve more than 94% of the secrecy rate performance with less than 1% computation time and more than 93% satisfaction of interference leakage constraints. These results are obtained with significant reduction in computational time, which we believe that it is suitable for future real-time wireless applications.

Automated Recognition of Retinal Pigment Epithelium Cells on Limited Training Samples Using Neural Networks.

PurposeTo develop a neural network (NN)–based approach, with limited training resources, that identifies and counts the number of retinal pigment epithelium (RPE) cells in confocal microscopy images obtained from cell culture or mice RPE/choroid flat-mounts.MethodsTraining and testing dataset contained two image types: wild-type mice RPE/choroid flat-mounts and ARPE 19 cells, stained for Rhodamine-phalloidin, and imaged with confocal microscopy. After image preprocessing for denoising and contrast adjustment, scale-invariant feature transform descriptors were used for feature extraction. Training labels were derived from cells in the original training images, annotated and converted to Gaussian density maps. NNs were trained using the set of training input features, such that the obtained NN models accurately predicted corresponding Gaussian density maps and thus accurately identifies/counts the cells in any such image.ResultsTraining and testing datasets contained 229 images from ARPE19 and 85 images from RPE/choroid flat-mounts. Within two data sets, 30% and 10% of the images, were selected for validation. We achieved 96.48% ± 6.56% and 96.88% ± 3.68% accuracy (95% CI), on ARPE19 and RPE/choroid flat-mounts.ConclusionsWe developed an NN-based approach that can accurately estimate the number of RPE cells contained in confocal images. Our method achieved high accuracy with limited training images, proved that it can be effectively used on images with unclear and curvy boundaries, and outperformed existing relevant methods by decreasing prediction error and variance.Translational RelevanceThis approach allows efficient and effective characterization of RPE pathology and furthermore allows the assessment of novel therapeutics.

Joint Pedestrian Motion State and Device Pose Classification

Novel features for joint classification of gait and device modes are proposed and multiple machine learning methods are adopted to jointly classify the modes. The classification accuracy as well as the F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> score of two standard classification algorithms, K-nearest neighbor (KNN) and Gaussian process (GP), are evaluated and compared against a proposed neural network (NN)-based classifier. The proposed features are the correlation scores of a detected gait cycle relative to a set of unique gait signatures as well as the gait cycle time, all extracted from handheld inertial measurement units (IMUs). Each gait signature is defined such that it contains one full cycle of the human gait. In order to take the temporal correlation between classes into account, the initial classifiers' estimates are fed into a hidden Markov model (HMM) unit to obtain the final class estimates. The performance of the proposed method is evaluated on a large data set, including two classes of gait modes (walking and running) and four classes of device modes (fixed and face-up in the hand, swinging in the hand, in the pocket, and in the backpack). The experimental results validate the reliability of the considered features and effectiveness of the HMM unit. The initial classification accuracy of the NN-based approach is 91%, which is further improved to 99% after the smoothing stage on the validation set and 98% on the test set.

Probe Selection and Power Weighting in Multiprobe OTA Testing: A Neural Network-Based Approach

Over-the-air (OTA) radiated testing is an efficient solution to evaluate the performance of multiple-input multiple-output (MIMO) capable devices, which can emulate realistic multipath channel conditions in a controlled manner within lab environment. In a multiprobe anechoic chamber- (MPAC-) based OTA setup, determining the most appropriate probe locations and their power weights is critical to improve the accuracy of channel emulation at reasonable system costs. In this paper, a novel approach based on neural networks (NNs) is proposed to derive suitable angular locations as well as power weights of OTA probe antennas; in particular, by using the regularization technique, active probe locations and their weights can be optimized simultaneously with only one training process of the proposed NN. Simulations demonstrate that compared with the convex optimization-based approach to perform probe selection in the literature, e.g., the well-known multishot algorithm, the proposed NN-based approach can yield similar channel emulation accuracy with significantly reduced computational complexity.

Cu Diffusion in Amorphous Ta2O5 Studied with a Simplified Neural Network Potential

Understanding atomistic details of diffusion processes in amorphous structures is becoming increasingly important due to the recent advances in various information and energy devices. Atomistic simulations based on the density functional theory (DFT) represent a powerful approach; however, the development of a method characterized by both high reliability and computational efficiency remains a challenge. In this study, a simple neural network (NN) interatomic potential is constructed from the results of DFT simulations to investigate the diffusion of a single Cu atom in amorphous Ta2O5. The proposed technique is as accurate as the DFT in predicting hopping paths and the corresponding barrier energies in a given amorphous structure. Although the developed NN-based approach exhibited some limitations since it was constructed specifically for Cu, the obtained results showed that the NN potential was able to satisfactorily describe the Cu diffusion behavior. Thus, the Cu diffusion activation energy calculated...

Synthetic retrieval of hourly net ecosystem exchange using the neural network model with combined MI and GOCI geostationary sensor datasets and ground-based measurements

ABSTRACTNet ecosystem carbon dioxide (CO2) exchange (NEE) is a key parameter for understanding the terrestrial plant ecosystems, but it is difficult to monitor or predict over large areas at fine temporal resolutions. In this research, we estimated the hourly NEE using a combination of the integrated neural network (NN) model with geostationary satellite imagery to overcome the limitations of existing daily polar orbiting satellite-derived carbon flux products. Two sets of satellite imageries (i.e. the meteorological imager (MI) and geostationary ocean colour imager (GOCI) aboard communication, ocean, and meteorological satellite (COMS)) and CO2 flux data derived from eddy covariance measurements were used to verify the feasibility of applying hourly geostationary satellite imagery with an NN-based approach for estimating NEE at high temporal resolutions. For the NN model, the optimum neuronal architecture was established using an NN with one hidden layer that was trained using the Levenberg–Marquardt back propagation algorithm. The hourly NEE values estimated in test period from the NN model using the combined COMS MI and GOCI imagery and ground measurements as model inputs were compared with the eddy covariance NEE values from the measurement tower, which yielded reliable statistical agreement. The hourly NEE results from the NN model based on COMS MI and GOCI imagery and ground measurement data had the highest accuracy (RMSE = 2.026 μmol m−2 s−2, R = 0.975), while the root mean square error (RMSE) and the regression coefficient (R) generated by the NN model based on satellite imagery as the sole input variable were relatively lower (RMSE = 3.230 μmol m−2 s−2, R = 0.952). Although the simulations for the satellite-only NEE were showed as lower accuracy than the NN model that included all input variables, the hourly variations in NEE also appeared to describe its daily growth and development pattern well, indicating the possibility of deriving hourly-based products from the proposed NN model using geostationary satellite data as inputs.