Search

Computer-aided diagnosis of ocular disorders like glaucoma can be effectively performed with retinal fundus images. Since the advent of machine learning and later deep learning techniques in medical image analysis, research focusing on automated and early detection of glaucoma has gained prominence. In this paper, we show how a deep Convolutional Neural Network model can be used in visualising the features related to Glaucoma that are consistent with (a subset of) those used by ophthalmologists for diagnosis. However, since the doctors do not necessarily depend entirely on these features but back them up with other important measurements such as visual field testing (HVF), any deep neural network such as CNN may not necessarily be expected to yield 100% accuracy. Further, other important parameters such as Intraocular Pressure (IOP), which may be built into a deep learning model, may not necessarily correlate well with the presence of Glaucoma. In this work, a deep CNN model has been trained and tested on a large number of high-quality fundus images, both normal and glaucomatous. We compare the results with transfer learning models such as VGG16, ResNet50, and MobileNetV2. On an average, we obtained an accuracy of 93.75% in identifying glaucoma, focusing only on the features of the fundus image. [ABSTRACT FROM AUTHOR]

This work presents an on-device machine learning model with the ability to identify different mobility gestures called human activity recognition (HAR), which includes running, walking, squatting, jumping, and others. The data is collected through an Arduino Nano 33 BLE Sense board with a sampling rate of 119 Hz, which is embedded with an Inertial Measurement Unit (IMU) sensor. The same board is used as a microcontroller to identify human gestures by developing an end-to-end edge computing application. A deep neural network model is trained and then compressed for deployment on the board to create a self-contained, embedded device capable of identifying the type of gesture performed. Three deep learning models, namely Multi-Layer Perceptron (MLP), Convolutional Neural Network – Long Short Term Memory (CNN-LSTM) & CNN-Gated Recurrent Unit (CNN-GRU), are evaluated in the identification of the mobility gestures. The observed accuracy of the models is 96%, 97.1% and 97.8%, respectively, MLP, CNN-LSTM & CNN-GRU across the different gesture categories. The study shows the utility of embedded devices with deep neural network-based models, which can provide low cost, minimal power usage, and meet data privacy requirements in HAR. [ABSTRACT FROM AUTHOR]

In deep learning, GANs (Generative Adversarial Networks) are one of the prominent study areas due to their ability to generate synthetic data thereby solving the problem of the unavailability and the limited data sets. GAN is a framework of deep neural networks that can learn from a set of training data and generate new data with similar characteristics as the training data. In this review, the history of GAN, various types of GANs, objective functions, loss functions, and performance analysis done for GANs in various fields are also analysed. The main objective of the paper is to analyse the application of GANs in face restoration, and medical imaging including their evaluation metrics and data sets used. A deep review has been carried out on various types of GANs, their architecture, objective functions, and applications. This review focuses more on the medical applications using various types of GANs including image augmentation, disease detection, medical image enhancement, face restoration, detection, etc. The challenges, current progress, and future applications using various GANs are also discussed. This review clearly shows that the application of GANs has increased considerably and thus it proves a promising future in the field of deep learning. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy (DR) is considered to be a leading cause of blindness in people aged 16 to 64 years, affecting around 40% of the population diagnosed with diabetes mellitus (DM). DR is usually identified through retinal image analysis, and in some countries, such as Brazil, such diagnosis is hindered by limited access to specialized care, leading to lengthy waits for ophthalmological evaluations. This scenario makes the Brazilian Diabetic Society's annual DR check recommendation impractical for many. To address this gap, our study introduces a novel Siamese Convolutional Neural Network (SCNN) for DR prediction, usable by primary care professionals. Our SCNN analyzes pairs of eye fundus images and employs shared weights in its layers to extract essential features, facilitating similarity measurement between neural network outputs. Despite challenges with limited and imbalanced datasets, our SCNN showed effectiveness. We tested it on four datasets (IDRiD, APTOS, Messidor-1, DIARETDB0), with accuracy ranging from 67.23% (APTOS) to 96.85% (DIARETDB0). Compared to other methods, our approach consistently excelled, especially in recall analysis. These results suggest that deep learning via Siamese networks is likely to be a viable and potential DR screening tool. [ABSTRACT FROM AUTHOR]

Blindness or loss of vision is primarily caused by diabetic retinopathy, or DR. It is a retinal condition that may cause harm to the eyes' blood vessels.Due to the complicated eye structure, manually detecting diabetic retinopathy takes time and is prone to human mistake. There have been several automatic methods proposed for the diagnosis of diabetic retinopathy from fundus images.Using a combination of InceptionV3 and ResNet50, an optimised deep Convolutional Neural Network model of feature fusion is presented in this paper to categories diabetic retinopathy photos into five kinds. Concatenating results in the suggested model including both the Inception Modules and the Residual Blocks, this mitigates the gradient problem and reduces the number of parameters. Resnet50 and Inceptionv3, two different deep learning (DL) models, were used. The proposed hybrid InceptionV3 and ResNet50 based optimised deep CNN (IR_ODCNN) model gathers features from the data and joins them before sending them to the DCNN that has been optimised for classification. The hyper parameters for CNN layer training are optimised using the Grey Wolf Optimizer (GWO) method. An image dataset of the fundus is used to assess the suggested model. The experimental findings indicate that, when compared to the current approaches, In terms of classifying diabetic retinopathy, the proposed model performs better. [ABSTRACT FROM AUTHOR]

It is imperative to understand the significance of the early diagnosis of pneumonia using a convolutional neural network (CNN) to reduce the processing time and increase the quality of treatment that is delivered to the patient. We have implemented transfer learning for processing of available datasets and constructed an ensemble of 3-CNNs: SqueezeNet, ResNet-50 and EfficientNet-b0. In this work, a multiclassification model has been built that can help in early pneumonia diagnosis. The chest X-rays of patients have been classified in 2-stages. In the first stage, the EfficientNet-b0 convolutional neural network with 99% accuracy is employed to diagnose whether the patient's chest X-ray is Normal/Abnormal. If the output of the first stage is found abnormal then the chest X-ray is processed to the second stage wherein ResNet-50 with 97% accuracy is employed to diagnose the pneumonia type, pneumonia bacteria/pneumonia virus. Further, performance metrics have been computed from the confusion matrix for both stages of X-ray. Also, it is imperative to mention that a pneumonia diagnosis app has been developed in Matlab-2023 for the ease of patients who can self-evaluate the scan report and understand the course of clinical treatment. [ABSTRACT FROM AUTHOR]

In this work, the segmented dental X-ray images obtained by dentists have been classified into ideal/minimally compromised edentulous area (no clinical treatment needed immediately), partially/moderately compromised edentulous area (require bridges or cast partial denture) and substantially compromised edentulous area (require complete denture prosthesis). A total of 116 image dental X-ray dataset is used, of which 70% of the image dataset is used for training the convolutional neural network (CNN) while 30% is used sfor testing and validation. Three pretrained deep neural networks (DNNs; SqueezeNet, ResNet-50 and EfficientNet-b0) have been implemented using Deep Network Designer module of Matlab 2022. Each of these CNNs were trained, tested and optimised for the best possible accuracy and validation of dental images, which require an appropriate clinical treatment. The highest classification accuracy of 98% was obtained for EfficientNet-b0. This novel research enables the implementation of DNN parameters for automated identification and labelling of edentulous area, which would require clinical treatment. Also, the performance metrics, accuracy, recall, precision and F1 score have been calculated for the best DNN using confusion matrix. [ABSTRACT FROM AUTHOR]

Epilepsy is a long-term neurological condition that disrupts brain function in people of all ages, epilepsy is a condition that is analysed through the brain signals via electroencephalogram (EEG) signal. To analyse epilepsy using spatial and temporal data, various machine-learning-based techniques are used. However, most of the techniques suffer from inaccuracy issues in dealing with the dynamic and raw EEG signal. In this research, an intelligent societal optimisation-driven classifier is introduced based on convolutional neural networks (CNN) for epileptic seizure prediction using EEG signals. To boost predictive accuracy, we extract frequency band features from the EEG signal utilising wavelet decomposition. The frequency band features form the feature vector, is provided smart societal optimisation- CNN such that the prediction performance is enhanced through the optimal tuning of the CNN with the smart societal optimisation. Smart societal optimisation is proposed by integrating the behaviour of the Lobos wolf and the Moggie. The smart societal optimisation-based CNN attains 87.673% accuracy, 84.949% sensitivity91.274%specificity for the K-Fold-10 for CHB-MIT scalp EEG database. [ABSTRACT FROM AUTHOR]

Nowadays, deep neural networks (DNNs) are helpful tools for mammogram classification in breast cancer screening. But, in Vietnam, there is a relatively small number of mammograms for training DNNs. Therefore, this study aims to apply transfer learning techniques to improve the performance of DNN models. In the first step, 10,418 breast cancer images from the Digital Database for Screening Mammography were used for training the CNN model ResNet 34. In the second step, we fine-tune this model on the Hanoi Medical University (HMU) database with 6,248 Vietnamese mammograms. The optimal model of ResNet 34 among these models achieves a macAUC of 0.766 in classifying breast cancer X-ray images into three Breast Imaging-Reporting and Data System (BI-RADS) categories, BI-RADS 045 ('incomplete and malignance'), BI-RADS 1 ('normal'), and BI-RADS 23 ('benign'), when tested on the test dataset. This result is higher than the result of the ResNet 50 model trained only on the X-ray dataset of 7,912 breast cancer images of the HMU dataset, which achieves a macAUC of 0.754. A comparison of the performance of the proposed model of ResNet 34 applying transfer learning with other works shows that our model's evaluation results are higher than those of the compared models. [ABSTRACT FROM AUTHOR]

Objectives: Colorectal cancer (CRC) is a prevalent disease characterized by significant dysregulation of gene expression. Non-invasive tests that utilize microRNAs (miRNAs) have shown promise for early CRC detection. This study aims to determine the association between miRNAs and key genes in CRC. Methods: Two datasets (GSE106817 and GSE23878) were extracted from the NCBI Gene Expression Omnibus database. Penalized logistic regression (PLR) and artificial neural networks (ANN) were used to identify relevant miRNAs and evaluate the classification accuracy of the selected miRNAs. The findings were validated through bipartite miRNA-mRNA interactions. Results: Our analysis identified 3 miRNAs: miR-1228, miR-6765-5p, and miR-6787-5p, achieving a total accuracy of over 90%. Based on the results of the mRNA-miRNA interaction network, CDK1 and MAD2L1 were identified as target genes of miR-6787-5p. Conclusions: Our results suggest that the identified miRNAs and target genes could serve as non-invasive biomarkers for diagnosing colorectal cancer, pending laboratory confirmation. [ABSTRACT FROM AUTHOR]

This study analyses the interplay between media framing and hate speech in South Korean refugee-related news. As news about refugees within Korea has recently emerged in the Korean media, analyzing how news framing influences public discourse would serve as the starting point for understanding the media's role in shaping public opinions towards refugees. Using 3965 news articles and 314,769 comments on Afghan and Yemeni refugees, we employed the analysis of topic model network to identify frames. Deep neural networks detected hate comments, and Poisson regression assessed the impact of framing and its interaction with news scope on hate speech. For Afghan refugees, international news with a victim frame reduced hate comment, whereas domestically it increased them. Intruder framing generally decreased hate comments, particularly in domestic news. Similarly, for Yemeni refugees, international victim-framed news received fewer hate comments, while domestic news with the same frame increased hate comments. The intruder frame, which had no effect on domestic news, significantly increased hateful comments on international news. These results show how nuances of framing, along with news scope, affect the dynamics of hate speech in South Korea when refugees are portrayed as a domestic or as an international issue. [ABSTRACT FROM AUTHOR]

Maintaining the stability of laser melt-pool conditions is critical to ensure consistency in cooling and solidification rates during directed energy deposition (DED) process. This serves to minimize degradation in the built accuracy and can improve overall microstructure homogeneity in DED parts. This work presents a novel laser power planning approach by utilizing a deep neural network (DNN) agent to learn the laser power control policy directly from the melt-pool information predicted by the finite element (FE) model. The agent was trained via the deep Q-learning algorithm using a previously developed FE model as the process simulator. The numerical experiment demonstrates that the trained agent is capable of modulating the laser power during the simulated deposition of a new component and can maintain the fluctuation in melt-pool volume within the defined target range. This work shows the potential of using the numerical modeling and reinforcement learning to generate the required laser power planning and modulation in DED process, without requiring physical sensors and closed-loop in-process monitoring and control. [ABSTRACT FROM AUTHOR]

Summary: Deep learning (DL)‐based channel estimation for orthogonal frequency division multiplexing with index modulation (OFDM‐IM) under Rayleigh fading channel conditions is presented in this paper. A deep neural network (DNN) is utilized to estimate the channel response in simulations. The proposed DNN is trained using the channel coefficient derived through the least squares (LS) method. Then channel estimation is conducted using the trained DNN. Within the DNN, the long short‐term memory (LSTM) layer is included as the hidden layer. Different scenarios are handled in simulations and the proposed DNN is compared with traditional channel estimation methods. The simulations demonstrate that the DL‐based channel estimation significantly surpasses the LS and minimum mean‐square error (MMSE) techniques. [ABSTRACT FROM AUTHOR]

Graph Neural Networks (GNNs) models, a current machine learning hotspot, have increasingly started to be applied in fraud detection in conjunction with user reviews in recent years. The accessible material is complicated and varied, the aggregated user evaluations cover a diverse range of topics, and erroneous information among vast amounts of user-generated content is typically rare. The review system is modeled as a heterogeneous network to address the issue of feature heterogeneity and uneven data distribution, and a new social theory-based graphical neural network model (SGNN) is suggested. The rich user behavior information in the heterogeneous network may be fully leveraged to acquire richer semantic representations for comments by integrating the hierarchical attention structure. Under the ensemble learning bagging framework, various distinct SGNN sub-models are combined. The sampling technique realizes the diversity aggregation of the base learners, which reduces the loss of useful information and improves the ability to identify bogus comments. According to testing results on real datasets from Amazon and YelpChi, the SGNN approach provides strong anomaly detection performance. It is demonstrated that the SGNN process has good robustness against fraudulent entities in the use of skewed distribution of data categories when compared to the existing approach. [ABSTRACT FROM AUTHOR]

The core part of this study is comprised of using the soft computational approach for analyzing the thermoelastic reaction of a concrete sandwich beam with transversely-graded composite face sheets as an advanced building material exposed to an axially-directed abrupt thermal load. First of all, the relations governing motion are extracted from the principles of three-dimensional elasticity theory. Besides, the time-variable relations of heat conduction are derived based on the Fourier heat conduction cooperating with Lord–Shulman theorem of heat transfer. Continuity of the displacements and stresses on the core's borders interacting with face sheets is taken into consideration. To find the solution of the time-variable governing equations of motion, the Laplace transform in conjunction with the differential quadrature approach (DQA) is utilized. Transforming the system's reaction from Laplace space to the time domain is done through an extended expression of Dubner and Abate's approach. To speed up the computational process, a soft computational approach on the bases of optimizing a deep neural network with a shuffled frog leaping algorithm (SFLA) is implemented. Efficiency of the applied methods is inspected though comparing their outcomes with those obtained from the existing literature for different boundary conditions. [ABSTRACT FROM AUTHOR]

In the present study, the machine learning based probabilistic model is proposed to perform the vibration analysis of functionally graded (FG) porous plates reinforced with graphene nanoplatelets (GNP). The plate is considered with the statistical variation in material properties such as elastic moduli of metal matrix and GNP, porosity index, and weight fraction of GNP while sampling via normal distribution. The effective material properties, i.e. elastic modulus and density of the GNP reinforced porous composite are respectively obtained using Halpin-Tsai and the rule of mixture models. The size and pore density are varied along the thickness direction using continuous cosine functions. The equation of motion based on higher-order shear deformable theory (HSDT) is derived using the variational principle which is further simulated by employing the finite element method (FEM) to obtain natural frequency. The database is generated based on selective input features and natural frequency i.e. output, and is further divided into 70% training dataset and 30% for testing and validation. A regression-based artificial neural network (ANN) model is adopted to learn the underlying function that maps the natural frequency to the inputs. The mapped model is further coupled with Monte Carlo simulation (MCS) to efficiently find the statistical variation in the natural frequency concerning variation in material properties (inputs) at reduced cost and time. [ABSTRACT FROM AUTHOR]

The dependencies between process parameters and the resulting geometrical accuracy of additively manufactured parts are usually highly non-linear and thus complex to investigate and mathematically quantify. Therefore, the application of artificial intelligence techniques is promising to generate mathematical models that reduce effort and increase the prediction quality. The overall goal is to establish a procedure to automatically determine the optimal settings of the manufacturing process parameters to guarantee the highest geometrical accuracy of parts in additive manufactured production. This paper presents the first step towards this fully automatic procedure – the training and evaluation of a mathematical model based on artificial neural networks to quantify the effects of varying process parameters of a material extrusion process on both macro- and micro-geometrical performances. Therefore, a dataset is established based on the Design of Experiment of an additively manufactured part made from Polylactic Acid filament. The dataset is then used to train an artificial neural network that predicts the dimensional and micro-geometrical deviations of the manufactured parts. Finally, the evaluation of the network's prediction quality and reliability indicate that it is possible to predict the parameters linked to resulting print quality with a mean absolute error from 0.0004 to 0.036. [ABSTRACT FROM AUTHOR]

Some of the viruses may cause lung parenchyma and airway involvement. Usually, viral pneumonia causes ground-glass opacities, bilateral peripheral distribution, consolidation, vascular thickening, and reticular opacity. These features are common in COVID-19 rather than Non-Covid-19 viral pneumonia. However, in advanced cases, COVID-19 viral pneumonia may cause organising pneumonia and fibrosis of the lung. Atypical findings of Non-Covid-19 pneumonia have included central peripheral distribution, pleural effusion, lymphadenopathy, nodules, tree-in-bud opacities, and pneumothorax. Therefore, differentiating Non-Covid-19 pneumonia from COVID-19 pneumonia at chest computed tomography (CT) is necessary. In that case, CT scans of the thorax are one of the essential tools for early identification and future prognosis of viral pneumonia. We have proposed a Computer-Aided Diagnostic (CAD) system that can detect features of chest CT using a Deep Neural Network (DNN) with Long Short-Term Memory (LSTM). Transfer learning using pre-trained DNN models (ResNet50, VGG19, InceptionV3, Xception, DenseNet121, and VGG16) is applied to retain both high-level and low-level features effectively. The deep features are passed to the LSTM layer. The LSTM is utilised as a classifier and detects long short-term dependencies. The proposed method employs a hybrid DNN-LSTM network for automatic detection to take advantage of the uniqueness of the two models. The proposed models are trained with common and different features present in the chest CT of COVID-19 and Non-Covid-19 viral pneumonia. The 5-fold cross-validation (CV) method validated and tested the proposed model. The proposed DNN model's performance is quite improved with LSTM and CV. As a result, the proposed LM-DNN (VGG16+LSTM+CV) model has achieved the classification test accuracy of 91.58% and specificity of 93.86%, which offers superior performance with state-of-the-art. Also, the DenseNet121+LSTM+CV model has reached the classification test accuracy of 90.1% and sensitivity of 92%. [ABSTRACT FROM AUTHOR]

Setting up a complex CNN requires a powerful platform, several hours of run-time, and a lot of data for training. Here, we propose a generalised lightweight solution that exploits super-resolution and scalable vector graphics and uses a small-scale UNet as the baseline framework to segment different organs in MR and CT data. We selected the UNet since many researchers use it as the baseline, modify it in their proposal, and perform an ablation study to show the effectiveness of the proposed modification. First, we downsample the input 2D CT slices by bicubic interpolation. Using the architecture of the conventional UNet, we reduce the size of the network's input, and the number of layers and filters to construct a lightweight UNet. The network segments the low-resolution images and prepares the mask of an organ. Then, we upscale the boundary of the output mask by the Scalable Vector Graphics technique to obtain the final border. This design reduces the number of parameters and the run-time by a factor of two. We segmented several tissues to prove the stability of our method to the type of organ. The experiments proved the feasibility of setting up complex deep neural networks with conventional platforms. [ABSTRACT FROM AUTHOR]

Recently, huge concerns have been raised in diagnosing chest diseases, especially after the COVID-19 pandemic. Regular diagnosis processes of chest diseases sometimes fail to distinguish between Corona and Viral Pneumonia diseases through Polymerase Chain Reaction (PCR) tests which are a time-engrossing process that needs convoluted manual procedures. Artificial Intelligence (AI) techniques have achieved high performance in aiding medical diagnostic processes. The innovation of this work lies in using a new diagnostic technique to distinguish between COVID-19 and Viral Pneumonia diseases using advanced AI technologies. This is done by extracting novel features from chest X-ray images based on Wavelet analysis, Scale Invariant Feature Transformation (SIFT), and the Mel Frequency Cepstral Coefficient (MFCC). Support vector machines (SVM) and artificial neural networks (ANN) were utilized to build classification algorithms using 1200 chest X-ray mages for each case. Using Wavelet features, the results of evaluating the SVM and ANN models were 97% accurate, and with SIFT features, they were closer to 99%. The proposed models were very effective at identifying COVID-19 and Viral Pneumonitis, so physicians can determine the best treatment course for patients with the support of this high accuracy. Moreover, this model can be used in hospitals and emergency rooms when a massive number of patients are waiting, as it is faster and more accurate than the regular diagnosis processes as each step takes few seconds on average to complete. [ABSTRACT FROM AUTHOR]

Despite their diversity and importance as ecological indicators and ecosystem service providers, the macroecology of bats in South Asia is poorly understood, and until recently studies on the ecological niches of these species have been rare. This study analyses the ecogeographic predictors of habitat suitability in South Asian bats by conducting ensemble ecological niche modelling using four algorithms (random forests, artificial neural networks, multivariate adaptive regression splines, and maximum entropy) to define suitability envelopes for 48 selected bat species, based on topographic, hydrographic, land-use, land-cover, and other anthropogenic impact factors. Anthropogenic impact variables showed high importance with Median Night-time Light being the biggest driver of habitat suitability for most of the study species with generally lower suitability of brighter areas. Projected suitable areas for individual species covered between 6.28% and 22.98% of the study area. Regions such as the Thar desert of northwestern India were consistently identified to have low suitability. The Western Ghats in India, the Himalayas in Bhutan, northern India, and Nepal, and Sri Lanka were identified as suitability hotspots for more than half the studied species overlapping with human-impacted habitats. This study offers insight into the impacts of anthropogenic pressure on the macroecology of bats in a megadiverse region and stresses the importance of analysing ecogeographic effects on ecological niches and habitat suitability, which can be vital to inform conservation planning and policymaking in the future. [ABSTRACT FROM AUTHOR]

The utilization of two-dimensional van der waals heterostructures in optoelectronic synapses allows for the integration of information processing and memory, thereby providing novel operating platforms for simulating the perceptual visual systems and developing the neuromorphic computing systems due to its contactless, highly efficient and parallel computing. Herein, we have constructed a straightforward MoS2/WS2 heterostructure optoelectronic synapse and examined its capacity to imitate synaptic behaviors under optical stimulus. The MoS2/WS2 device demonstrated several synaptic functions, such as the excitatory postsynaptic current, short-term plasticity, long-term plasticity, pairs-pulse facilitation and 'learning-experience' behavior. Moreover, the MoS2/WS2 synaptic device can achieve a wide range of photo response wavelengths, spanning from UV to visible light, as well as the conversion from short-term plasticity to long-term plasticity. Furthermore, light-induced charge transfer due to adsorption and desorption of oxygen molecules in MoS2/WS2 heterostructure can be used to explain its working mechanism. Additionally, the synaptic plasticity of MoS2/WS2 device can be controlled by adjusting the duration, power and number of the optical pulses, which renders the MoS2/WS2-based optoelectronic synaptic device extremely favorable for implementation in the perceptual visual system. [ABSTRACT FROM AUTHOR]

In this research, a novel design stochastic numerical technique is presented to investigate the unsteady form magnetohydrodynamic (MHD) slip flow along the boundary layer to analyze the transportation and heat transfer in a solar collector through nano liquids which is a revolution in the field of neurocomputing. Thermal conductivity in variable form is dependent on temperature and wall slips are assumed over the boundary. For mathematical modeling, the solar collector is assumed in the form of a nonlinear stretching sheet and a quite new artificial neural networks (ANNs) based approach is used to solve the current problem in which inverse multiquadric radial basis (IMRB) kernel is sandwiched between a global search solver named genetic algorithms (GAs) and a highly effective local solver named sequential quadratic programming (SQP) i.e. IMRB-GASQP solver. The governing boundary value problem is altered in the form of a system of nonlinear ordinary differential equations (ODEs) through the utilization of similarity transformation and then the obtained system of ODEs is solved using IMRB-GASQP solver by altering the values of distinguished parameters involved in it to observe the fluctuation in the velocity and temperature profiles of nanofluid. The obtained results are effectively compared with the reference solutions using the Adams numerical technique in graphical and tabulated form. An exhaustive error analysis using performance operators is presented while the efficacy of the designed solver using various statistical operators is also part of this research. [ABSTRACT FROM AUTHOR]

We use a neural network model and Atacama Large Millimeter/submillimeter Array (ALMA) observations of HCN and HNC to constrain the physical conditions, most notably the cosmic-ray ionization rate (CRIR, ζ), in the Central Molecular Zone (CMZ) of the starburst galaxy NGC 253. Using output from the chemical code UCLCHEM, we train a neural network model to emulate UCLCHEM and derive HCN and HNC molecular abundances from a given set of physical conditions. We combine the neural network with radiative transfer modeling to generate modeled integrated intensities, which we compare to measurements of HCN and HNC from the ALMA Large Program ALCHEMI. Using a Bayesian nested sampling framework, we constrain the CRIR, molecular gas volume and column densities, kinetic temperature, and beam-filling factor across NGC 253's CMZ. The neural network model successfully recovers UCLCHEM molecular abundances with ∼3% error and, when used with our Bayesian inference algorithm, increases the parameter-inference speed tenfold. We create images of these physical parameters across NGC 253's CMZ at 50 pc resolution and find that the CRIR, in addition to the other gas parameters, is spatially variable with ζ ∼ a few ×10−14 s−1 at r ≳ 100 pc from the nucleus, increasing to ζ > 10−13 s−1 at its center. These inferred CRIRs are consistent within 1 dex with theoretical predictions based on nonthermal emission. Additionally, the high CRIRs estimated in NGC 253's CMZ can be explained by the large number of cosmic-ray-producing sources as well as a potential suppression of cosmic-ray diffusion near their injection sites. [ABSTRACT FROM AUTHOR]

Machine Learning (ML) has emerged as a pivotal force in enhancing Proton Exchange Membrane Water Electrolyzer (PEMWE) devices. These devices are critical for transforming renewable electricity into hydrogen, a key clean energy vector. Despite their prospects, the broader implementation of PEMWE is hindered by cost and efficiency barriers. PEMWEs are inherently complex, involving multi-scale processes such as electrochemical reactions, reactant transportation, and thermo-electrical interactions. This complexity has previously limited optimizations to isolated components like electrocatalysts, membrane electrode assemblies (MEAs), Bipolar plates (BPs), and Gas Diffusion Electrodes (GDEs). ML presents a revolutionary pathway to address these obstacles by enabling system-wide optimization. In this paper, we offer an in-depth review of cutting-edge ML applications for improving PEMWE performance and efficiency. ML's ability to process large datasets and identify intricate patterns accelerates the research and development of PEMWEs, thereby reducing costs and boosting efficiency. We describe a variety of algorithms, such as Artificial Neural Networks (ANN), Deep Learning (DL), Long Short-Term Memory (LSTM), Support Vector Machine (SVM), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM), commonly used in PEMWE applications, highlighting their significance in enhancing PEMWE systems. Additionally, we explore hybrid methods that combine various ML techniques to further improve PEMWE performance and efficiency. The review provides a concise overview and forward-looking perspective on the role of ML in advancing PEMWE technology, marking a significant step towards their cost-effective and scalable deployment. • A thorough evaluation of PEMWE components and their specific functions is presented. • A wide range of ML algorithms for optimizing PEMWE are summarized. • Recent advancements in ML models for the design and optimization of PEMWE are reviewed. • The main challenges and future research opportunities in ML-driven PEMWE are discussed. [ABSTRACT FROM AUTHOR]

Finding an accurate model is essential for classification of respiratory pathologies through extraction and fusion of respiratory sounds' features. To handle the unlabeled data, a sequence of autoencoders are used for data augmentation. A deep neural network framework is proposed to extract three types of features: 1-Features based on the human auditory system, using Mel-frequency cepstral coefficients (MFCCs), 2-Temporal features contained in the sequence of sound signal using the long short-term memory (LSTM) network, and 3-The nonlinear and complex relationship among temporal characteristics in neighborhood regions using a two-dimensional (2D) convolutional neural network (CNN) applied to the sound sequence converted to a 2D time array. Three branches are fused through some fully connected layers. The ICBHI 2017 sound database is used in two cases of 6 and 3 pathological classes. In the case of 6-classes database, 98.72% overall accuracy, 98.46% kappa coefficient, 99.66% sensitivity, 98.70% specificity and 99.70% F1-Sscore is provided. In the case of 3-classes database, 97.18% overall accuracy, 95.77% kappa coefficient, 98.31% sensitivity, 99.16% specificity and 98.93% F1-score are achieved. In contrast to the many researchers that are applying CNN on the Mel spectrogram of the audios, CNN is applied to the 2D form of the time series to find the hidden relationship among the time instants in local regions. Moreover, fusion of these hidden features with the LSTM and MFCC features leads to an accurate multiple classification. This work proposes a new deep learning framework for fusion of three types of sound features, which leads to a significant improvement in multiple respiratory infection diagnosing. [ABSTRACT FROM AUTHOR]

X-ray Bragg coherent diffraction imaging is a powerful technique for 3D materials characterization. However, obtaining X-ray diffraction data is difficult and computationally intensive, motivating the need for automated processing of coherent diffraction images, with the goal of minimizing the number of X-ray datasets needed. We automate a machine learning approach to identify crystalline line defects in samples from the raw coherent diffraction data, in a workflow coupling coherent diffraction data generation with training and inference of deep neural network defect classifiers. In particular, we adopt a continual learning approach, where we generate training data as needed based on the accuracy of the defect classifier instead of generating all training data a priori. Moreover, we develop a novel data generation mechanism to improve the efficiency of defect identification beyond the previously published continual learning approach. We call the improved method smart continual learning. The results show that our approach improves the accuracy of defect classifiers and reduces training data requirements by up to 98% compared with prior approaches. [ABSTRACT FROM AUTHOR]

In this paper, a novel methodology is developed for the characterization of the capacity of rectangular-shaped concrete-filled steel tubes (CFSTs). In the scientific research field, of particular interest is the behavior of long CFST columns under eccentric compressive load. These conditions promote failure mechanisms involving global member buckling. The developed methodologies are based on machine learning techniques found on artificial neural networks (ANNs). Furthermore, optimization methodologies, employing the grey wolf optimization algorithm and the firefly algorithm, have been attempted. For the training and validation of the models, a database consisting of 1,641 experimental tests collected from literature sources has been prepared, containing long and short specimens as well as specimens with or without load eccentricity. As the vast majority of the available experimental tests involve short specimens, the database has been augmented with 216 3D finite element models (FEMs), featuring increased member slenderness values. The calibration of the FEMs has been performed against experimental tests. The performance of the developed models has been measured through a number of performance indices, and compared with available code procedures. They have been found to provide significant improvements, both for short and long CFST columns, with the ANN model optimized with the firefly algorithm outperforming the others. Furthermore, a graphical user interface (GUI) has been developed which can be readily used to estimate the axial load capacity of CFST columns through the optimal ANN model. The developed GUI is made available as a supplementary material. [ABSTRACT FROM AUTHOR]

This paper presents a new strategy that uses multiple neural networks in conjunction with the DEtection TRansformer (DETR) network to detect firearms in surveillance images. The strategy developed in this work presents a methodology that promotes collaboration and self-coordination of networks in the fully connected layers of DETR through the technique of multiple self-coordinating artificial neural networks (MANN), which does not require a coordinator. This self-coordination consists of training the networks one after the other and integrating their outputs without an extra element called a coordinator. The results indicate that the proposed network is highly effective, achieving high-level outcomes in firearm detection. The network's high precision of 84% and its ability to perform classifications are noteworthy. [ABSTRACT FROM AUTHOR]

Stochastic optimizers play a crucial role in the successful training of deep neural network models. To achieve optimal model performance, designers must carefully select both model and optimizer hyperparameters. However, this process is frequently demanding in terms of computational resources and processing time. While it is a well-established practice to tune the entire set of optimizer hyperparameters for peak performance, there is still a lack of clarity regarding the individual influence of hyperparameters mislabeled as "low priority", including the safeguard factor ϵ and decay rate β , in leading adaptive stochastic optimizers like the Adam optimizer. In this manuscript, we introduce a new framework based on the empirical probability density function of the loss' gradient magnitude, termed as the "gradient magnitude histogram", for a thorough analysis of adaptive stochastic optimizers and the safeguard hyperparameter ϵ . This framework reveals and justifies valuable relationships and dependencies among hyperparameters in connection to optimal performance across diverse tasks, such as classification, language modeling and machine translation. Furthermore, we propose a novel algorithm using gradient magnitude histograms to automatically estimate a refined and accurate search space for the optimal safeguard hyperparameter ϵ , surpassing the conventional trial-and-error methodology by establishing a worst-case search space that is two times narrower. [ABSTRACT FROM AUTHOR]

An accelerated failure time (AFT) model assumes a log‐linear relationship between failure times and a set of covariates. In contrast to other popular survival models that work on hazard functions, the effects of covariates are directly on failure times, the interpretation of which is intuitive. The semiparametric AFT model that does not specify the error distribution is sufficiently flexible and robust to depart from the distributional assumption. Owing to its desirable features, this class of model has been considered a promising alternative to the popular Cox model in the analysis of censored failure time data. However, in these AFT models, a linear predictor for the mean is typically assumed. Little research has addressed the non‐linearity of predictors when modeling the mean. Deep neural networks (DNNs) have received much attention over the past few decades and have achieved remarkable success in a variety of fields. DNNs have a number of notable advantages and have been shown to be particularly useful in addressing non‐linearity. Here, we propose applying a DNN to fit AFT models using Gehan‐type loss combined with a sub‐sampling technique. Finite sample properties of the proposed DNN and rank‐based AFT model (DeepR‐AFT) were investigated via an extensive simulation study. The DeepR‐AFT model showed superior performance over its parametric and semiparametric counterparts when the predictor was nonlinear. For linear predictors, DeepR‐AFT performed better when the dimensions of the covariates were large. The superior performance of the proposed DeepR‐AFT was demonstrated using three real datasets. [ABSTRACT FROM AUTHOR]

Although the anatomical arrangement of brain regions and the functional structures within them are similar across individuals, the representation of neural information, such as recorded brain activity, varies among individuals owing to various factors. Therefore, appropriate conversion and translation of brain information is essential when decoding neural information using a model trained using another person's data or to achieving brain-to-brain communication. We propose a brain representation transfer method that involves transforming a data representation obtained from one person's brain into that obtained from another person's brain, without relying on corresponding label information between the transferred datasets. We defined the requirements to enable such brain representation transfer and developed an algorithm that distills the assumption of common similarity structure across the brain datasets into a rotational and reflectional transformation across low-dimensional hyperspheres using encoders for non-linear dimensional reduction. We first validated our proposed method using data from artificial neural networks as substitute neural activity and examining various experimental factors. We then evaluated the applicability of our method to real brain activity using functional magnetic resonance imaging response data acquired from human participants. The results of these validation experiments showed that our method successfully performed representation transfer and achieved transformations in some cases that were similar to those obtained when using corresponding label information. Additionally, we reconstructed images from individuals' data without training personalized decoders by performing brain representation transfer. The results suggest that our unsupervised transfer method is useful for the reapplication of existing models personalized to specific participants and datasets to decode brain information from other individuals. Our findings also serve as a proof of concept for the methodology, enabling the exchange of the latent properties of neural information representing individuals' sensations. [ABSTRACT FROM AUTHOR]

Machine learning is playing a crucial role in optimizing material synthesis, particularly in scenarios where several parameters related to growth exhibit different and significant outcomes. An example of such complexity is the growth of atomically thin semiconductors through chemical vapor deposition (CVD), where multiple parameters can influence the thermodynamics and reaction kinetics involved in the synthesis. Herein, we performed a set of orthogonal experiments, varying the key parameters such as temperature, carries gas flux and precursor position to identify the optimal conditions for maximizing covered area and the size of rhenium disulfide (ReS2) crystals. The experimental results were used to establish correlations among the three thermodynamic variables through an artificial neural network. Contour plots were then generated to visualize the impact on the coverage and flake size of the crystals. This study demonstrates the capability of machine learning to enhance the potential of CVD-growth for the integration of 2D semiconductors like ReS2 at larger scales. [ABSTRACT FROM AUTHOR]

Highlights Cocured bonding and geometric modification are prominent techniques for fusing structural components in aerospace and aircraft applications. This research involved the fabrication of carbon fiber‐reinforced polymeric (CFRP) adhesive‐bonded joints utilizing novel cocuring with interleaved lamination (IL‐CC) joining techniques with multiwalled carbon nanotubes (MWCNTs) modified epoxy adhesive. Additionally, the study examined the flexural and vibration characteristics of IL‐CC CFRP composite joints. The result findings 1.0 wt% MWCNTs in epoxy adhesive had the best flexural strength and modulus, 78% and 32% higher than the pure epoxy adhesive. Additionally, the IL‐CC CFRP composite joints exhibit a 285% higher flexural strength and a 47% higher modulus than the neat CC CFRP composite joints. Nevertheless, the nanofiller with a wt% of 0.25 demonstrated the highest natural frequency across three different vibration modes, which are 14%, 21%, and 25% higher than pure epoxy adhesive. An ANOVA showed that MWCNT concentrations significantly influenced performance. Using the Levenberg–Marquardt algorithm, an artificial neural network predicted results. The overall coefficient (R) mean square error of 0.99627 is satisfactory, indicating that both outcomes are dependable and in good agreement. The results imply that IL‐CC CFRP composite joints could be used in aerospace and aircraft parts. The IL‐CC CFRP SLJs were fabricated with MWCNT‐modified adhesive. Attained the maximum flexural strength is 1.0 wt% MWCNT. Attained the maximum natural frequency is 0.25 wt% MWCNT. Statistics were analyzed using a one‐way ANOVA technique. An ANN predicts the flexural and dynamic behavior of IL‐CC CFRP SLJs. [ABSTRACT FROM AUTHOR]

The high binding affinity of antibodies toward their cognate targets is key to eliciting effective immune responses, as well as to the use of antibodies as research and therapeutic tools. Here, we propose ANTIPASTI, a convolutional neural network model that achieves state-of-the-art performance in the prediction of antibody binding affinity using as input a representation of antibody-antigen structures in terms of normal mode correlation maps derived from elastic network models. This representation captures not only structural features but energetic patterns of local and global residue fluctuations. The learnt representations are interpretable: they reveal similarities of binding patterns among antibodies targeting the same antigen type, and can be used to quantify the importance of antibody regions contributing to binding affinity. Our results show the importance of the antigen imprint in the normal mode landscape, and the dominance of cooperative effects and long-range correlations between antibody regions to determine binding affinity. [Display omitted] • Antibody-antigen binding affinity prediction using normal modes and deep learning • ANTIPASTI achieves state-of-the-art accuracy and generalization power • ANTIPASTI is interpretable and identifies key antibody regions for binding • Both short and long-range structural correlations are crucial for binding affinity Understanding the drivers of antibody-antigen binding affinity is crucial to design optimized antibodies for research and therapeutic purposes. Michalewicz et al. present ANTIPASTI, a deep learning model that predicts binding affinity from structural data and identifies which antibody regions are important for binding. [ABSTRACT FROM AUTHOR]

In this work, we present the neural network Mesh Exploring Tensor Net (METNet) for the segmentation of 3D urban scenes, that operates directly on textured meshes. Since triangular meshes have a very irregular structure, many existing approaches change the input by sampling evenly distributed point clouds on the meshes. The resulting simplified representation of the urban scenes has the advantage that state-of-the-art neural network architectures for point clouds can be used for the segmentation task. The disadvantages are that crucial geodesic information is lost and for a sufficiently good approximation of the scene many points are often necessary. Since memory on the GPU is limited, the consequence is that only small areas can be examined locally. To overcome these limitations, METNet generates its input directly from the textured triangular mesh. It applies a new star-shaped exploration strategy, starting from a triangle on the mesh and expanding in various directions. This way, a vertex based regular local tensor from the unstructured mesh is generated, which we call a Mesh Exploring Tensor (MET). By expanding on the mesh, a MET maintains the information about the connectivity and distance of vertices along the surface of the mesh. It also effectively captures the characteristics of large regions. Our new architecture, METNet is optimized for processing METs as input. The regular structure of the input allows METNet the use of established convolutional neural networks. Experimental results, conducted on two urban textured mesh benchmarks, demonstrate that METNet surpasses the performance of previous state-of-the-art techniques. METNet improves the previous state-of-the-art results by 8.6% in terms of mean IoU (intersection over union) on the SUM dataset compared to the second-best method PSSNet, and by 3.2% in terms of mean F1 score on the H3D dataset compared to the second-best method ifp-SCN. Our source code is available at https://github.com/QenSchr/METNet. [ABSTRACT FROM AUTHOR]

How do newborns learn to see? We propose that visual systems are space-time fitters, meaning visual development can be understood as a blind fitting process (akin to evolution) in which visual systems gradually adapt to the spatiotemporal data distributions in the newborn's environment. To test whether space-time fitting is a viable theory for learning how to see, we performed parallel controlled-rearing experiments on newborn chicks and deep neural networks (DNNs), including CNNs and transformers. First, we raised newborn chicks in impoverished environments containing a single object, then simulated those environments in a video game engine. Second, we recorded first-person images from agents moving through the virtual animal chambers and used those images to train DNNs. Third, we compared the viewpoint-invariant object recognition performance of the chicks and DNNs. When DNNs received the same visual diet (training data) as chicks, the models developed common object recognition skills as chicks. DNNs that used time as a teaching signal—space-time fitters—also showed common patterns of successes and failures across the test viewpoints as chicks. Thus, DNNs can learn object recognition in the same impoverished environments as newborn animals. We argue that space-time fitters can serve as formal scientific models of newborn visual systems, providing image-computable models for studying how newborns learn to see from raw visual experiences. Author summary: Do machines learn like brains? The performance of all learning systems depends on both the learning machinery and experiences (training data) from which the system learns, so answering this question will require giving machines and brains the same training data. To do so, we introduce a digital twin method for running parallel controlled-rearing studies on newborn animals and deep neural networks. We show that when deep neural networks (CNNs and transformers) are trained in the same visual environments as newborn chicks, the models develop the same object recognition skills as chicks. Both newborn chicks and deep neural networks can learn invariant object representations that generalize across novel viewpoints, even when learning occurs in an impoverished environment containing a single object seen from a limited 60° viewpoint range. Our study shows that blind fitting processes (variation + selection learning) can mimic the rapid visual learning of precocial newborn animals, in the absence of innate (hardcoded) knowledge about objects or space. We argue that visual development can be understood as space-time fitting, in which visual systems gradually adapt to the spatiotemporal data distributions in the environment. [ABSTRACT FROM AUTHOR]

ABSTRACT Traditional physical models and purely data‐driven approaches often struggle with small sample sizes and the complex effects of strain ratios. To overcome these challenges, this study integrates physical principles with machine learning techniques to improve fatigue life predictions for natural rubber (NR). A uniaxial fatigue test on NR was performed, generating data to construct a physical model. A physics‐informed neural network (PINN) model was subsequently developed, utilizing the fatigue life predicted by the physical model, along with engineering strain amplitude and strain ratio as input variables, whereas the experimentally observed fatigue life served as the output variable. The accuracy of the physical model, a data‐driven model, and the proposed PINN model was evaluated by comparing their predictions against measured fatigue life data. The findings demonstrate that the PINN model significantly enhances prediction accuracy, with its fatigue life estimates consistently falling within 1.5 times the measured values. [ABSTRACT FROM AUTHOR]

Highlights In this study, the impact performance of carbon fiber‐reinforced polymer (CFRP) composites under low‐speed impact conditions was investigated using a data‐driven approach. Both material properties and impact parameters were determined through experimental methods and finite element (FE) analysis. FE analysis was conducted on CFRP composite structures to generate impact force and absorbed energy datasets. Various impact conditions, such as impactor height (0.5–1.25 m), impactor shape (flat, truncated cone, bullet, and cone), and composite plate thickness (1–4 mm), were incorporated into an artificial neural network (ANN) model to predict the impact behavior of CFRP composites. Using the optimal plate thickness identified from the data‐driven model, CFRP plates were fabricated using vacuum‐assisted resin transfer molding and tested under different impactor shapes and heights using drop impactors. The FE analysis revealed that increasing the impactor height improved the impact force by 37.8% and the absorbed energy by 178%. The impactor shape also significantly influenced the results, with a flat‐to‐cone‐shaped impactor increasing the impact force and absorbed energy by 167.2% and 440%, respectively. Additionally, the plate thickness analysis showed that a 2 mm plate provided optimal impact force and absorbed energy, with values of 1.09 kN and 8.12 J, respectively. The prediction of the force and energy experienced by CFRP material under different impact conditions was validated using root mean squared error (RMSE), mean square error (MSE), mean absolute error (MAE), and R2 metrics. The model demonstrated excellent performance with the lowest RMSE (0.0118), MSE (0.0003), and MAE (0.0096), indicating that the predicted impact forces closely matched the actual forces. The highest R2 value (0.9999) suggests that the model accurately captures the variance in impact force across varied impact conditions. Similarly, R2 values close to one indicate that the model effectively explains the variability in energy absorption, making it highly reliable. The ANN model also showed that predictions for absorbed energy were more accurate than those for impact force under varying impact conditions. Furthermore, the predicted impact force and absorbed energy from the FE analysis and ANN model closely aligned with the experimental results. Damage morphology observations indicated matrix cracking at higher impact velocities, with more significant penetration occurring with cone‐shaped impactors. These findings demonstrate a strong correlation between experimental and numerical outcomes, validating the effectiveness of this combined approach for evaluating the impact resistance of CFRP composites. The impact performance of CFRP composites under different impact conditions was modeled. An Artificial Neural Network model was developed to predict impact performance. The VARTM method was adopted for the development of optimized CFRP composites. Impactor height and shape were found to be the most influential factors Impact damage areas were related based on impactor height and shape. [ABSTRACT FROM AUTHOR]

As the core device of vacuum circuit breaker, the insulation performance of the interrupter is closely related to the internal structure. In this paper, the electric field distribution of arc interrupter under simulated lightning impulse overvoltage is studied. It is found that the instantaneous electric field intensity is at a high level and the electric field uniformity is poor under lightning impulse. In order to further improve the electric field distribution of the interrupter, this paper proposes a method based on the combination of neural network and particle swarm optimization (PSO) to optimize the important structural parameters of the interrupter. A neural network model was established with suspension shield length (L1), suspension shield radius (R1), end shield radius (R2), contact plate thickness (H1) and end shield length (H2) as inputs, and the maximum electric field strength and the uniformity of electric field on both sides of the static and dynamic contacts as outputs, respectively. The structure of the vacuum interrupter was optimized by PSO algorithm. The optimization results show that: When the structural parameters of the vacuum interrupter H1 are 4.6 mm, H2 is 20 mm, R1 is 3.6 mm, R2 is 3.4 mm and L1 is 110 mm, the electric field distribution of the interrupter is obviously improved, and the electric field distribution of the optimized structure can also be well improved under the power frequency operating voltage. Its electric field intensity between the electrode is significantly low. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Determining the location of objects relative to ourselves is essential for interacting with the world. Neural activity in the retina is used to form a vision-independent model of the local spatial environment relative to the body. For example, when an animal navigates through a forest, it rapidly shifts its gaze to identify the position of important objects, such as a tree obstructing its path. This seemingly trivial behavior belies a sophisticated neural computation. Visual information entering the brain in a retinocentric reference frame must be transformed into an egocentric reference frame to guide motor planning and action. This, in turn, allows the animal to extract the location of the tree and plan a path around it. In this review, we explore the anatomical, physiological, and computational implementation of retinocentric-to-egocentric reference frame transformations — a research area undergoing rapid progress stimulated by an ever-expanding molecular, physiological, and computational toolbox for probing neural circuits. We begin by summarizing evidence for retinocentric and egocentric reference frames in the brains of diverse organisms, from insects to primates. Next, we cover how distance estimation contributes to creating a three-dimensional representation of local space. We then review proposed implementations of reference frame transformations across different biological and artificial neural networks. Finally, we discuss how an internal egocentric model of the environment is maintained independently of the sensory inputs from which it is derived. By comparing findings across a variety of nervous systems and behaviors, we aim to inspire new avenues for investigating the neural basis of reference frame transformation, a canonical computation critical for modeling the external environment and guiding goal-directed behavior. Visual input reaches the brain in retina-centered coordinates but is transformed into new reference frames to support spatial cognition and behavioral outputs. Martins et al. review the neural circuit and computational implementations of this transformation in the nervous systems of diverse animal species and discuss future avenues for experimental investigation. [ABSTRACT FROM AUTHOR]

The relationship between chemical structure and chiroptical properties is not always clearly understood. Nowadays, efforts to develop new systems with enhanced optical properties follow the trial‐error method. A large number of data would allow us to obtain more robust conclusions and guide research toward molecules with practical applications. In this sense, in this work we predict the chiroptical properties of millions of halogenated [6]helicenes in terms of the rotatory strength (R). We have used DFT calculations to randomly create derivatives including from 1 to 16 halogen atoms, that were then used as a data set to train different deep neural network models. These models allow us to i) predict the Rmax for any halogenated [6]helicene with a very low computational cost, and ii) to understand the physical reasons that favour some substitutions over others. Finally, we synthesized derivatives with higher predicted Rmax obtaining excellent correlation among the values obtained experimentally and the predicted ones. [ABSTRACT FROM AUTHOR]

This paper develops a method for predicting the atomization energy Eat of multi-atomic compounds by representing the molecular configuration as a sorted Coulomb matrix M(Ri, Zi). A dataset of 7165 molecules with H, C, N, O, S content is analyzed and the input data for training is generated. The optimal layer configuration and hyperparameters values were found for the fully-connected neural network model. The minimum mean absolute error MAE ≈ 7.66 kcal/mol was achieved in predicting the atomization energy Eat. [ABSTRACT FROM AUTHOR]

Prediction of pressure evolution in liquid hydrogen (LH 2) tanks utilizing data-driven technology has gradually garnered significant attention. However, training an accurate data-driven model for pressure prediction in LH 2 tanks is a challenge due to inadequate experimental data. To end this, this paper presents a Backpropagation Neural Network model for pressure prediction in LH 2 tanks based on transfer learning on pre-trained models, using experimental data and results from thermodynamics models. Pre-trained models are firstly optimized by Bayesian optimization algorithms with large sample datasets from the thermal multi-zone model program and then performed the fine-tuning method with small sample datasets from experiments. The effectiveness of the proposed approach is validated by experimental data and numerical results. Compared with baseline models trained only with experimental data, the results demonstrate better model performance, with a maximum of 10.51% accuracy improvement. The proposed approach demonstrates considerable potential in facilitating the application of advanced machine learning algorithms for LH 2 tank pressure prediction, significantly reducing the dependence on experimental data collection. [Display omitted] • Data-driven pressure prediction model in liquid hydrogen tanks. • Data augmentation by integrating both experimental and simulation data. • Finding the optimal model by Bayesian optimization algorithms. • A maximum of 10.51% improvement in accuracy after transfer learning. [ABSTRACT FROM AUTHOR]

Speaker-conditioned target speaker extraction aims at estimating the target speaker from a mixture of speakers utilizing auxiliary information about the target speaker. In this paper, we consider a single-channel target speaker extraction system consisting of a speaker embedder network and a speaker separator network. Instead of using standard long short-term memory (LSTM) cells in the separator network, we propose two variants of LSTM cells that are customized for speaker-conditioned target speaker extraction. The first variant customizes both the forget gate and input gate of the LSTM cell, aiming at retaining only relevant features related to target speaker and disregarding the interfering speakers by simultaneously resetting and updating the cell state using the speaker embedding. For the second variant, we introduce a new gate within the LSTM cell, referred to as auxiliary-modulation gate. This gate modulates the information processing during cell state reset, aiming at learning the long-term and short-term discriminative features of the target speaker. Both in unidirectional and bidirectional mode, experimental results on 2-speaker mixtures, 3-speaker mixtures, and noisy mixtures (containing 1, 2, or 3 speakers) show that both proposed variants of LSTM cells outperform the standard LSTM cells for target speaker extraction, where the best performance is obtained using the auxiliary-gated LSTM cells. [ABSTRACT FROM AUTHOR]

In this study, the short-beam shear strength (SBSS) retention of two types of glass fiber-reinforced polymer (GFRP) bars—sand-coated (SG) and ribbed (RG)—was subjected to alkaline, acidic, and water conditions for up to 12 months under both high-temperature and ambient laboratory conditions. Comparative assessments were also performed on older-generation sand-coated (SG-O) and ribbed (RG-O1 and RG-O2) GFRP bars exposed to identical conditions. The results demonstrate that the new-generation GFRP bars, SG and RG, exhibited significantly better durability in harsh environments and exhibited SBSS retentions varying from 61 to 100% in SG and 90–98% in RG under the harshest conditions compared to 56–69% in SG-O, 71–80% in RG-O1, and 74–88% in RG-O2. Additionally, predictive models using both artificial neural networks (ANNs) and linear regression were developed to estimate the strength retention. The ANN model, with an R2 of 0.95, outperformed the linear regression model (R2 = 0.76), highlighting its greater accuracy and suitability for predicting the SBSS of GFRP bars. [ABSTRACT FROM AUTHOR]

Background: Accurate prediction of neurorehabilitation outcomes following Spinal Cord Injury (SCI) is crucial for optimizing healthcare resource allocation and improving rehabilitation strategies. Artificial Neural Networks (ANNs) may identify complex prognostic factors in patients with SCI. However, the influence of psychological variables on rehabilitation outcomes remains underexplored despite their potential impact on recovery success. Methods: A cohort of 303 patients with SCI was analyzed with an ANN model that employed 17 input variables, structured into two hidden layers and a single output node. Clinical and psychological data were integrated to predict functional outcomes, which were measured by the Spinal Cord Independence Measure (SCIM) at discharge. Paired Wilcoxon tests were used to evaluate pre–post differences and linear regression was used to assess correlations, with Pearson's coefficient and the Root Mean Square Error calculated. Results: Significant improvements in SCIM scores were observed (21.8 ± 15.8 at admission vs. 57.4 ± 22.5 at discharge, p < 0.001). The model assigned the highest predictive weight to SCIM at admission (10.3%), while psychological factors accounted for 36.3%, increasing to 40.9% in traumatic SCI cases. Anxiety and depression were the most influential psychological predictors. The correlation between the predicted and actual SCIM scores was R = 0.794 for the entire sample and R = 0.940 for traumatic cases. Conclusions: The ANN model demonstrated the strong impact, especially for traumatic SCI, of psychological factors on functional outcomes. Anxiety and depression emerged as dominant negative predictors. Conversely, self-esteem and emotional regulation functioned as protective factors increasing functional outcomes. These findings support the integration of psychological assessments into predictive models to enhance accuracy in SCI rehabilitation outcomes. [ABSTRACT FROM AUTHOR]

We propose data-driven models based on artificial neural networks (ANN) to predict changes in water-oil relative permeability curves given a salinity reduction in the injection water. The ANN consisted of a multilayer feedforward structure with backpropagation. For validation, a database from a semi-empirical correlation was created, and models with added noise were used to analyze the influence of the data dispersion. Then, a survey of experimental relative permeability curves was performed to produce a real database for sandstone and carbonate rocks, utilized in the training of the final models, with hyperparameter optimization and cross-validation. The initial model was able to consistently reproduce the original correlation, with a mean squared error (MSE) on the order of 10 - 6 . In the noise-trained model, the error measured was lower than the analytical error expected from random dispersion. In models trained with real data, the adopted strategy led to a final training MSE on the order of 10 - 3 , with better performance in networks with two hidden layers. The obtained models are useful in modeling relative permeabilities for low-salinity and engineered water injection projects. Training can be continuously updated with new data, and the methodology can be applied to other properties or even other multivariate regression problems. [ABSTRACT FROM AUTHOR]

The performance of photovoltaic solar panels is influenced by their temperature, so there is a need for a tool that can accurately and instantly predict the panel temperature. This paper presents an analysis of the panel temperature's dependence on atmospheric parameters at an operational photovoltaic park in the semi-arid north of Chile using Artificial Neural Networks (ANNs). We applied the back-propagation algorithm to train the model by using the atmospheric variables tilted solar radiation (TSR), air temperature, and wind speed measured in the park. The ANN model's effectiveness was evaluated by comparing it to five different deterministic models: the Standard model, King's model, Faiman's model, Mattei's model, and Skoplaki's model. Additionally, we examined the sensitivity of panel temperature to changes in air temperature, TSR, and wind speed. Our findings show that the ANN model had the best prediction accuracy for panel temperature, with a Root Mean Squared Error (RMSE) of 1.59 °C, followed by Mattei's model with a higher RMSE of 3.30 °C. We also determined that air temperature has the most significant impact on panel temperature, followed by TSR and wind speed. These results demonstrate that the ANN is a powerful tool for predicting panel temperature with high accuracy. [ABSTRACT FROM AUTHOR]

Deep learning methods have been applied when working to enhance the prediction accuracy of traditional statistical methods in the field of plant breeding. Although deep learning seems to be a promising approach for genomic prediction, it has proven to have some limitations, since its conventional methods fail to leverage all available information. Multimodal deep learning methods aim to improve the predictive power of their unimodal counterparts by introducing several modalities (sources) of input information. In this review, we introduce some theoretical basic concepts of multimodal deep learning and provide a list of the most widely used neural network architectures in deep learning, as well as the available strategies to fuse data from different modalities. We mention some of the available computational resources for the practical implementation of multimodal deep learning problems. We finally performed a review of applications of multimodal deep learning to genomic selection in plant breeding and other related fields. We present a meta-picture of the practical performance of multimodal deep learning methods to highlight how these tools can help address complex problems in the field of plant breeding. We discussed some relevant considerations that researchers should keep in mind when applying multimodal deep learning methods. Multimodal deep learning holds significant potential for various fields, including genomic selection. While multimodal deep learning displays enhanced prediction capabilities over unimodal deep learning and other machine learning methods, it demands more computational resources. Multimodal deep learning effectively captures intermodal interactions, especially when integrating data from different sources. To apply multimodal deep learning in genomic selection, suitable architectures and fusion strategies must be chosen. It is relevant to keep in mind that multimodal deep learning, like unimodal deep learning, is a powerful tool but should be carefully applied. Given its predictive edge over traditional methods, multimodal deep learning is valuable in addressing challenges in plant breeding and food security amid a growing global population. [ABSTRACT FROM AUTHOR]