High Performance Metrics Research Articles

Seabed classification of multibeam echosounder data into bedrock/non-bedrock using deep learning

The accurate mapping of seafloor substrate types plays a major role in understanding the distribution of benthic marine communities and planning a sustainable exploitation of marine resources. Traditionally, this activity has relied on the efforts of marine geology experts, who accomplish it manually by examining information from acoustic data along with the available ground-truth samples. However, this approach is challenging and time-consuming. Hence, it is important to explore automatic methods to replace this manual process. In this study, we investigated the potential of deep learning (U-Net) for classifying the seabed as either “bedrock” or “non-bedrock” using bathymetry and/or backscatter data, acquired with multibeam echosounders (MBES). Slope and hillshade data, derived from the bathymetry, were also included in the experiment. Several U-Net models, taking as input either one of these datasets or a combination of them, were trained using an expert delineated map as reference. The analysis revealed that U-Net has the ability to map bedrock and non-bedrock areas reliably. On our test set, the models using either bathymetry or slope data showed the highest performance metrics and the best visual match with the reference map. We also observed that they often identified topographically rough features as bedrock, which were not interpreted as such by the human expert. While such discrepancy would typically be considered an error of the model, the scale of the expert annotations as well as the different methods used by the experts to manually generate maps must be considered when evaluating the predictions quality. While encouraging results were obtained here, further research is necessary to explore the potential of deep learning in mapping other seabed types and evaluating the models’ generalization capabilities on similar datasets but different geographical locations.

Breast tumor prediction and feature importance score finding using machine learning algorithms

The subject matter of this study is breast tumor prediction and feature importance score finding using machine learning algorithms. The goal of this study was to develop an accurate predictive model for identifying breast tumors and determining the importance of various features in the prediction process. The tasks undertaken included collecting and preprocessing the Wisconsin Breast Cancer original dataset (WBCD). Dividing the dataset into training and testing sets, training using machine learning algorithms such as Random Forest, Decision Tree (DT), Logistic Regression, Multi-Layer Perceptron, Gradient Boosting Classifier, Gradient Boosting Classifier (GBC), and K-Nearest Neighbors, evaluating the models using performance metrics, and calculating feature importance scores. The methods used involve data collection, preprocessing, model training, and evaluation. The outcomes showed that the Random Forest model is the most reliable predictor with 98.56 % accuracy. A total of 699 instances were found, and 461 instances were reached using data optimization methods. In addition, we ranked the top features from the dataset by feature importance scores to determine how they affect the classification models. Furthermore, it was subjected to a 10-fold cross-validation process for performance analysis and comparison. The conclusions drawn from this study highlight the effectiveness of machine learning algorithms in breast tumor prediction, achieving high accuracy and robust performance metrics. In addition, the analysis of feature importance scores provides valuable insights into the key indicators of breast cancer development. These findings contribute to the field of breast cancer diagnosis and prediction by enhancing early detection and personalized treatment strategies and improving patient outcomes.

Dynamic parametric MRI and deep learning: Unveiling renal pathophysiology through accurate kidney size quantification.

Renal pathologies often manifest as alterations in kidney size, providing a valuable avenue for employing dynamic parametric MRI as a means to derive kidney size measurements for the diagnosis, treatment, and monitoring of renal disease. Furthermore, this approach holds significant potential in supporting MRI data-driven preclinical investigations into the intricate mechanisms underlying renal pathophysiology. The integration of deep learning algorithms is crucial in achieving rapid and precise segmentation of the kidney from temporally resolved parametric MRI, facilitating the use of kidney size as a meaningful (pre)clinical biomarker for renal disease. To explore this potential, we employed dynamic parametric T2 mapping of the kidney in rats in conjunction with a custom-tailored deep dilated U-Net (DDU-Net) architecture. The architecture was trained, validated, and tested on manually segmented ground truth kidney data, with benchmarking against an analytical segmentation model and a self-configuring no new U-Net. Subsequently, we applied our approach to in vivo longitudinal MRI data, incorporating interventions that emulate clinically relevant scenarios in rats. Our approach achieved high performance metrics, including a Dice coefficient of 0.98, coefficient of determination of 0.92, and a mean absolute percentage error of 1.1% compared with ground truth. The DDU-Net enabled automated and accurate quantification of acute changes in kidney size, such as aortic occlusion (-8% ± 1%), venous occlusion (5% ± 1%), furosemide administration (2% ± 1%), hypoxemia (-2% ± 1%), and contrast agent-induced acute kidney injury (11% ± 1%). This approach can potentially be instrumental for the development of dynamic parametric MRI-based tools for kidney disorders, offering unparalleled insights into renal pathophysiology.

Comparative Performance Analysis of Ensemble Learning Algorithms for Rock Classification

Knowing the physical and mechanical properties of rocks is important for engineering studies. Because determining the properties and type of rocks affects the safety of engineering structures. Automatic detection of rock types reduces the workload of engineers. In this study, the types of rocks were determined by using some physical and mechanical properties of rocks measured in the laboratory. Rep tree algorithm and ensemble learning algorithms were used in the study. The success of ensemble learning algorithms in classification was compared. As a result, it was understood that ensemble learning algorithms increase success. The most successful algorithm in rock classification was the Logistboost algorithm. The highest performance metrics were obtained in the classification made with the Logistboost algorithm. In addition, 4 different metric types were calculated to determine the error rates of the algorithms. Logistboost algorithm classified with the lowest error rate.

Improving Semantic Segmentation Performance in Underwater Images

Nowadays, semantic segmentation is used increasingly often in exploration by underwater robots. For example, it is used in autonomous navigation so that the robot can recognise the elements of its environment during the mission to avoid collisions. Other applications include the search for archaeological artefacts, the inspection of underwater structures or in species monitoring. Therefore, it is necessary to improve the performance in these tasks as much as possible. To this end, we compare some methods for image quality improvement and data augmentation and test whether higher performance metrics can be achieved with both strategies. The experiments are performed with the SegNet implementation and the SUIM dataset with eight common underwater classes to compare the obtained results with the already known ones. The results obtained with both strategies show that they are beneficial and lead to better performance results by achieving a mean IoU of 56% and an increased overall accuracy of 81.8%. The result for the individual classes shows that there are five classes with an IoU value close to 60% and only one class with an IoU value less than 30%, which is a more reliable result and is easier to use in real contexts.

A Boosted Evolutionary Neural Architecture Search for Timeseries Forecasting with Application to South African COVID-19 Cases

In recent years, there has been an increase in studies on time-series forecasting for the future occurrence of disease incidents. Improvements in deep learning approaches offer techniques for modelling long-term temporal relationships. Nonetheless, this design practice is rigorously painstaking, prone to errors, and requires human expertise. The advent of feature enrichment with automatic architecture search typically optimises the discovery of new neural architectures applicable in domains such as time-series modelling. The main methodological contribution of this study is an approach for time-series forecasting using feature-enriched filters and an evolutionary neural architecture search with sequence-to-sequence gated recurrent units (GRU-Seq2Seq). This is applied to the prediction of daily cases of coronavirus disease in South Africa. The highly pathogenic coronavirus pandemic incident data was modelled with filters, optimised hyper-parameter search trials and an evolutional neural algorithm. The proposed model was benchmarked against ARIMA and SARIMA. The model predicted trends for 30, 60 and 90-day horizons and evaluated them for 7, 14 and 31 days. Simulation results demonstrate that observed daily case counts with added filters and evolutionary search optimisation for forecasting improve performance accuracy. Generally, the proposed bFilter+GRU-Seq2Seq with optimal search configuration outperformed ARIMA and SARIMA with lower error scores and higher performance metrics, with an R2 score of 7.48E-01 for a 30-day forecast horizon.

Transition metal nitrides embedded in N-doped porous graphitic Carbon: Applications as electrocatalytic sulfur host materials

While transition metal nitrides (TMNs) are promising electrocatalysts, their widespread use is challenged by the complex synthetic methodology and a limited understanding of the underlying electrocatalytic mechanisms. Herein, we describe a novel synthesis of TMNs (including Mo2N, NbN, and ZrN) and explore their potential as electrocatalysts to affect sulfur cathode reactions. The TMNs were prepared in-situ using a process that simultaneously infuses nitrogen-doped porous graphitic carbon (designated as TMN@N-PGC). The methodology avoids the use of ammonia, which poses safety risks due to its flammability and toxicity. Analysis of the d-p hybridized orbitals formed between the transition metal ions and sulfur species revealed that the antibonding orbitals are empty. Thus, TMNs with more negative d-band centers exhibit stronger affinities towards polysulfides. NbN facilitated polysulfide conversion as well as Li2S detachment, and thus featured a high electrocatalytic capability for promoting cathode kinetics. Lithium−sulfur (Li–S) batteries containing NbN@N-PGC exhibited the highest performance metrics in terms of specific capacity (1488 mA h g−1 at 0.1 C), rate capacity (521 mA h g−1 at 6 C), and cycling stability (603 mA h g−1 at 0.5 C after 1300 cycles, corresponding a capacity decay of 0.030% per cycle). Li−S cells with high sulfur loadings also exhibit outstanding performance.

Target-Aware Feature Bottleneck for Real-Time Visual Tracking

Recent Siamese network-based visual tracking approaches have achieved high performance metrics on numerous recent visual tracking benchmarks, where most of these trackers employ a backbone feature extractor network with a prediction head network for classification and regression tasks. However, there has been a constant trend of employing a larger and complex backbone network and prediction head networks for improved performance, where increased computational load can slow down the overall speed of the tracking algorithm. To address the aforementioned issues, we propose a novel target-aware feature bottleneck module for trackers, where the proposed bottleneck can elicit a target-aware feature in order to obtain a compact feature representation from the backbone network for improved speed and robustness. Our lightweight target-aware bottleneck module attends to the feature representation of the target region to elicit scene-specific information and generate feature-wise modulation weights that can adaptively change the importance of each feature. The proposed tracker is evaluated on large-scale visual tracking datasets, GOT-10k and LaSOT, and we achieve real-time speed in terms of computation and obtain improved accuracy over the baseline tracker algorithm with high performance metrics.

Automated Detection and Measurement of Dural Sack Cross-Sectional Area in Lumbar Spine MRI Using Deep Learning.

Lumbar spine magnetic resonance imaging (MRI) is a critical diagnostic tool for the assessment of various spinal pathologies, including degenerative disc disease, spinal stenosis, and spondylolisthesis. The accurate identification and quantification of the dural sack cross-sectional area are essential for the evaluation of these conditions. Current manual measurement methods are time-consuming and prone to inter-observer variability. Our study developed and validated deep learning models, specifically U-Net, Attention U-Net, and MultiResUNet, for the automated detection and measurement of the dural sack area in lumbar spine MRI, using a dataset of 515 patients with symptomatic back pain and externally validating the results based on 50 patient scans. The U-Net model achieved an accuracy of 0.9990 and 0.9987 on the initial and external validation datasets, respectively. The Attention U-Net model reported an accuracy of 0.9992 and 0.9989, while the MultiResUNet model displayed a remarkable accuracy of 0.9996 and 0.9995, respectively. All models showed promising precision, recall, and F1-score metrics, along with reduced mean absolute errors compared to the ground truth manual method. In conclusion, our study demonstrates the potential of these deep learning models for the automated detection and measurement of the dural sack cross-sectional area in lumbar spine MRI. The proposed models achieve high-performance metrics in both the initial and external validation datasets, indicating their potential utility as valuable clinical tools for the evaluation of lumbar spine pathologies. Future studies with larger sample sizes and multicenter data are warranted to validate the generalizability of the model further and to explore the potential integration of this approach into routine clinical practice.

Femtosecond-Laser Microstructuring Improves Solid-State Electrolyte/Cathode Ionic Transport

All-solid-state batteries (ASSBs) provide the opportunity for substantially enhanced gravimetric and volumetric energy and power density and improved safety. However, these gains will not be realized without first overcoming substantial challenges, notably, the high interfacial resistance between electrodes and the solid-state electrolyte (SSE) which requires high stack pressures, on the order of 10,000 psi, to achieve high performance metrics.[1,2]. Already there is a large and growing body of research demonstrating the advantages of laser structuring of Li-ion battery materials for enhanced kinetics and shorter diffusion pathways [3,4]. This work extends these advantages to ASSBs through lowering the catholyte-SSE interfacial impedance by increasing its surface area and providing less torturous paths for ion diffusion.The work presented here uses a near-IR (1030 nm) femtosecond laser (pulse duration approx. 250 fs) to ablate microstructures into the surface of a slurry-cast LiNi0.8Mn0.1Co0.1O2 (NMC811):Li6PS5Cl catholyte. Micrometer-sized channel structures were ablated at various depths and pitch spacings to assess the effect of patterning on cell performance. X-ray diffraction (XRD) and energy dispersive spectroscopy (EDS) measurements were acquired on both pristine and patterned samples to ensure that the femtosecond laser pulses do not cause any residual compositional or phase changes, respectively. Critically, low interfacial resistance will require that an SSE layer be applied to the patterned catholyte such that the microstructure is preserved while maintaining intimate contact between the catholyte and SSE layer. SSE layers were both slurry cast and pellet-pressed onto both pristine and patterned electrodes. Cross-sectional SEM was used to investigate the degree of interfacial contact, and relate interface quality to parameters such as SSE application technique (e.g., pellet pressing vs. slurry casting) and slurry properties (i.e., viscosity). The improvement to interfacial ionic conductivity is assessed using electrochemical impedance spectroscopy (EIS). The effect of stack pressure is analyzed in order to assess the degree to which laser patterning can alleviate stack pressure requirements in ASSBs.

Uranium-based electrode material for the high-performance supercapacitors

In recent decades, significant progress has been made in the engineering of electrode materials for supercapacitors. In this regard, the primary emphasis has been on investigating carbonaceous based materials. However, this focus has restricted the overall capacitance of supercapacitor devices, thus hindering their broader application possibilities. There is still a gap in the literature on high-performance radioactive element-based electrode materials. Bearing this, in this work, a simple approach was suggested for the implementation of uranium, as a radioactive element, decorating onto carbon architecture, as electrode material in high-energy supercapacitor cells. Herein, a facile pathway for tailoring a uranium-decorated graphene oxide nanostructure (U-GOx; where x stands for the GO:U ratio by weight) was proposed and the potential applications of the synthesized nanoarchitectures as an electrode material of supercapacitor cell were investigated. As per the author's knowledge, for the first time in the ever-burgeoning field of energy storage systems, a radioactive compound-based supercapacitor cell was fabricated. The physicochemical characterizations confirmed the successful anchoring of uranium atoms onto the graphene oxide skeleton. The U-GO5 electrode-based-supercapacitor cell offered the highest performance metrics with a 95.92 % capacitance retention value at the end of the 15,000 consecutive galvanostatic charge-discharge cycles, besides the high specific capacitance value of 258.6 F⸱g−1 at a high current density value of 10.0 A⸱g−1. Moreover, the radioactivity measurements were conducted to investigate whether the fabricated supercapacitors were radioactive or not. The results revealed that the measured radiation dose of the U-GO5-based supercapacitor cell before the GCD cycles, which was of the highest uranium content, was ca.0.06 μSv (the background measurement was excluded). This finding indicated there would be no adverse effect on human health or the environment. It can be speculated that this work is a new era for engineering a high-performance supercapacitor based on organic electrodes decorated with a radioactive element.

CNN Approach for Predicting Survival Outcome of Patients With COVID-19

Coronavirus disease 2019 (COVID-19) has been challenged specifically with the new variant. The number of patients seeking treatment has increased significantly, putting tremendous pressure on hospitals and healthcare systems. With the potential of artificial intelligence (AI) to leverage clinicians to improve personalized medicine for COVID-19, we propose a deep learning model based on 1D and 3D convolutional neural networks (CNNs) to predict the survival outcome of COVID-19 patients. Our model consists of two CNN channels that operate with CT scans and the corresponding clinical variables. Specifically, each patient data set consists of CT images and the corresponding 44 clinical variables used in the 3D CNN and 1D CNN input, respectively. This model aims to combine imaging and clinical features to predict short-term from long-term survival. Our models demonstrate higher performance metrics compared to state-of-the-art models with AUC-ROC of 91.44 – 91.60% versus 84.36 – 88.10% and Accuracy of 83.39 – 84.47% versus 79.06 – 81.94% in predicting the survival groups of patients with COVID-19. Based on the findings, the combined clinical and imaging features in the deep CNN model can be used as a prognostic tool and help to distinguish censored and uncensored cases of COVID-19.

Millimeter-wave planar antenna array augmented with a low-cost 3D printed dielectric polarizer for sensing and internet of things (IoT) applications

A unique high gain antenna array with a 3D-printed dielectric polarizer is proposed. The packaging of the antenna array feeding structure is eliminated by aggregating the feeding network in between the antenna elements. This has a significant advantage in maintaining neat and symmetric radiation characteristics with low cross-polarization levels. The proposed structure combines two elements in one feeding point to reduce the array distribution feeding points of a 4 × 4 antenna array from 16 to 8 points. The proposed antenna array structure is extremely low in cost and can be used as either a linearly or circularly polarized one. The antenna array achieves a gain of 20 dBi/dBiC in both scenarios. The matching bandwidth is 4.1%, and the 3-dB Axial Ratio (AR) bandwidth is 6%. The antenna array uses a single substrate layer without the need for any vias. The proposed antenna array suits well various applications at 24 GHz, while maintaining high performance metrics, and low cost. The antenna array can be easily integrated with transceivers due to the use of printed microstrip line technology.

Automated Multimodal Stress Detection in Computer Office Workspace

Nowadays, changes in the conditions and nature of the workplace make it imperative to create unobtrusive systems for the automatic detection of occupational stress, which can be feasibly addressed through the adoption of Internet of Things (IoT) technologies and advances in data analysis. This paper presents the development of a multimodal automated stress detection system in an office environment that utilizes measurements derived from individuals’ interactions with the computer and its peripheral units. In our analysis, behavioral parameters of computer keyboard and mouse dynamics are combined with physiological parameters recorded by sensors embedded in a custom-made smart computer mouse device. To validate the system, we designed and implemented an experimental protocol simulating an office environment and included the most known work stressors. We applied known classifiers and different data labeling methods to the physiological and behavioral parameters extracted from the collected data, resulting in high-performance metrics. The feature-level fusion analysis of physiological and behavioral parameters successfully detected stress with an accuracy of 90.06% and F1 score of 0.90. The decision-level fusion analysis, combining the features extracted from both the computer mouse and keyboard, showed an average accuracy of 66% and an average F1 score of 0.56.

The Effect of Hyperparameter Optimization on the Estimation of Performance Metrics in Network Traffic Prediction using the Gradient Boosting Machine Model

Information and Communication Technology (ICT) has changed the way we communicate and access information, resulting in the high generation of heterogeneous data. The amount of network traffic generated constantly increases in velocity, veracity, and volume as we enter the era of big data. Network traffic classification and intrusion detection are very important for the early detection and identification of unnecessary network traffic. The Machine Learning (ML) approach has recently entered the center stage in network traffic accurate classification. However, in most cases, it does not apply model hyperparameter optimization. In this study, gradient boosting machine prediction was used with different hyperparameter optimization configurations, such as interaction depth, tree number, learning rate, and sampling. Data were collected through an experimental setup by using the Sophos firewall and Cisco router data loggers. Data analysis was conducted with R software version 4.2.0 with Rstudio Integrated Development Environment. The dataset was split into two partitions, where 70% was used for training the model and 30% for testing. At a learning rate of 0.1, interaction depth of 14, and tree number of 2500, the model estimated the highest performance metrics with an accuracy of 0.93 and R of 0.87 compared to 0.90 and 0.85 before model optimization. The same configuration attained the minimum classification error of 0.07 than 0.10 before model optimization. After model tweaking, a method was developed for achieving improved accuracy, R square, mean decrease in Gini coefficients for more than 8 features, lower classification error, root mean square error, logarithmic loss, and mean square error in the model.

Development and validation of explainable machine-learning models for carotid atherosclerosis early screening

BackgroundCarotid atherosclerosis (CAS), an important factor in the development of stroke, is a major public health concern. The aim of this study was to establish and validate machine learning (ML) models for early screening of CAS using routine health check-up indicators in northeast China.MethodsA total of 69,601 health check-up records from the health examination center of the First Hospital of China Medical University (Shenyang, China) were collected between 2018 and 2019. For the 2019 records, 80% were assigned to the training set and 20% to the testing set. The 2018 records were used as the external validation dataset. Ten ML algorithms, including decision tree (DT), K-nearest neighbors (KNN), logistic regression (LR), naive Bayes (NB), random forest (RF), multiplayer perceptron (MLP), extreme gradient boosting machine (XGB), gradient boosting decision tree (GBDT), linear support vector machine (SVM-linear), and non-linear support vector machine (SVM-nonlinear), were used to construct CAS screening models. The area under the receiver operating characteristic curve (auROC) and precision-recall curve (auPR) were used as measures of model performance. The SHapley Additive exPlanations (SHAP) method was used to demonstrate the interpretability of the optimal model.ResultsA total of 6315 records of patients undergoing carotid ultrasonography were collected; of these, 1632, 407, and 1141 patients were diagnosed with CAS in the training, internal validation, and external validation datasets, respectively. The GBDT model achieved the highest performance metrics with auROC of 0.860 (95% CI 0.839–0.880) in the internal validation dataset and 0.851 (95% CI 0.837–0.863) in the external validation dataset. Individuals with diabetes or those over 65 years of age showed low negative predictive value. In the interpretability analysis, age was the most important factor influencing the performance of the GBDT model, followed by sex and non-high-density lipoprotein cholesterol.ConclusionsThe ML models developed could provide good performance for CAS identification using routine health check-up indicators and could hopefully be applied in scenarios without ethnic and geographic heterogeneity for CAS prevention.

Free-form optimization of pattern shape for improving mechanical characteristics of a concentric tube

Medical concentric tubes have recently been designed using auxetic structures with negative Poisson's ratios to reduce undesirable snapping instability and improve mechanical performance. Despite the vast range of stiffness levels, prior auxetic designs were restricted to relatively simple geometries like square or triangular holes. Here, we present a novel free-form optimization for designing the bending and torsional stiffness of tubular structures using the multi-objective Bayesian optimization (MBO) method. The initial dataset of hole shapes is generated using a non-parametric Bézier curve, and the corresponding mechanical properties are estimated via numerical analysis. The acquisition function suggests a new hole shape with higher performance metrics based on the dataset, and updates to the regression model are made with new data points to improve prediction accuracy. Compared to a conventional tube with rectangular holes, the optimized design improves key factor, a ratio of bending stiffness to torsional rigidity, and torsional stiffness by 32% and 25%, respectively, by expanding the existing design space. The performance of the optimally designed tube was validated by resonant vibration tests.

Impedance-Matched Differential Superconducting Nanowire Detectors

Superconducting nanowire single-photon detectors (SNSPDs) are the highest-performing photon-counting technology in the near-infrared region. Because of delay-line effects, large-area SNSPDs typically trade off timing resolution and detection efficiency. This unavoidable fundamental constraint might limit their future deployment in demanding scientific applications. Here we introduce a detector design based on transmission-line engineering and differential readout for device-level signal conditioning, enabling a high system detection efficiency and a low detector jitter simultaneously. To make our differential detectors compatible with single-ended time taggers, we also engineer analog differential-to-single-ended readout electronics, with minimal impact on the system timing resolution. Our best niobium nitride differential SNSPD achieves a system detection efficiency of (83.3±4.3)% at 1550nm and (78±5)% at 775nm. The lowest system jitter is 13.0±0.4ps at 1550nm and 9.7±0.4ps at 775nm, limited by intrinsic contributions. These detectors also achieve sub-100-ps timing response at 1/100 of the maximum level, 30.7±0.4ps at 775 nm and 47.6±0.4ps at 1550nm, enabling time-correlated single-photon counting with high-dynamic-range response functions. Furthermore, because of the differential impedance-matched design, our detectors exhibit delay-line imaging capabilities and photon-number resolution. The properties and high-performance metrics achieved by our system make it a versatile photon-detection solution for quantum computing, quantum communication, and many other scientific applications.8 MoreReceived 16 December 2021Revised 7 February 2023Accepted 20 March 2023DOI:https://doi.org/10.1103/PhysRevApplied.19.044093© 2023 American Physical SocietyPhysics Subject Headings (PhySH)Research AreasOptoelectronicsPhysical SystemsLow-temperature superconductorsNanowiresOptical sources & detectorsSuperconducting devicesTechniquesInfrared techniquesPhoton countingSingle-photon detectorsQuantum InformationCondensed Matter, Materials & Applied PhysicsAtomic, Molecular & Optical

Convolutional neural network guided prediction of saturation magnetic flux density of Fe-based metallic glasses

Fe-based metallic glasses (FeMGs) have recently garnered considerable research attention due to their unique characteristics, such as low core loss and high saturation magnetic flux density (Bs). They also have promising applications in modern electrical machines, such as inductors, motors, and transformers. For FeMGs with high Bs, a trial-and-error approach using various metalloid elements as constituents has been intensively carried out to consider satisfactory thermal stability. However, this approach is time-consuming and expensive because it requires a huge compositional space and results in a lack of understanding of metalloid effects on magnetic properties and thermal stability. In this study, we established a deep learning model using a convolutional neural network (CNN) based only on compositional descriptors of FeMG constituent elements to predict Bs as the representative magnetic property. In addition, an artificial neural network was applied with the same compositional descriptors for comparison, resulting in a lower prediction performance than that of the CNN model. The CNN model demonstrated high-performance metrics that included a determination coefficient (R2) of 0.960 and a low root-mean-square error of 0.067 T, indicating that the material composition-to-property map is well-defined. We further applied the CNN model to unseen datasets for Bs prediction and compared it with results reported in the existing literature. The predicted Bs values are in excellent agreement with the reported experimental results.

Prediction of mechanical properties of high-entropy ceramics by deep learning with compositional descriptors

High-entropy ceramics (HECs) have garnered considerable research attention due to their unique characteristics and potential applications in extreme environmental conditions. As numerous elements can act as constituent materials for HECs, the trial-and-error approach as well as the computational approach are time-consuming and expensive for developing new HECs. In this study, we developed a deep learning model using a convolutional neural network (CNN) based on only the compositional descriptors of the constituent elements, e.g., rock-salt carbides, nitrides, and carbonitrides, to predict the corresponding mechanical properties of bulk (K), shear (G) and Young (Y) moduli. For comparison, XGboost, one of the ensemble algorithms, and an artificial neural network (ANN) were applied to predict the mechanical properties of HECs. The CNN models demonstrated the highest performance metrics of determination coefficient (R2 =0.856, 0.909, and 0.921 for K, G, and Y, respectively) and low root–mean–square errors (RMSE = 19.129, 14.540, and 31.929 GPa for K, G, and Y, respectively), which were consistent with the experimental results reported in previous studies. We further employed the CNN models to high-hardness HECs, evaluated their hardness (H) based on the empirical formulae of Tian, Chen, and Teter’s models, and compared them with the results reported in the existing literature. The hardness calculated using Teter’s model displayed excellent agreement with the experimental results for higher hardness HECs (H ≥ 27 GPa), while the estimated values by Tian’s models are consistent with the experimental results for lower hardness HECs (H < 27 GPa).