Model Configuration Research Articles

Investigating the thermodynamic optimization of naturally convective flow in a corrugated enclosure: The influence of gap spacing and orientation of split baffles

The natural convection in cavities is frequently used in fluid mechanics and heat transfer engineering, such as heat exchangers, electronics, solar collectors, and growing crystals. However, the physics of natural convection flow and heat transfer in cavities with split baffles is least understood. The fundamental aim of this research is to investigate the impact of heated split baffles positioned at various locations on steady-state free convection within a sinusoidal corrugated star cavity. In this model configuration, the outer wavy enclosure is maintained at a constant temperature of Tc, while the inner split baffles are heated at a constant temperature of Th. The finite element method is employed to discretize and solve the governing equations describing the fluid flow and heat transfer within the enclosure. This numerical approach aimed to analyze the effects of baffle inclination angles, baffle spacing, Rayleigh number on the fluid dynamics and convective thermal transport characteristics. The variation in velocity and temperature profile is illustrated through the streamlines and isotherm contours. Moreover, the numerical result is displayed in term Nuavg of the heat transfer, which are analyzed for inside space of baffles and angles of the baffle (θ=00,450,900). The key finding demonstrates that increasing the Rayleigh numbers and the different positions (up, central, down) of inner vertical split baffles enhances the magnitude of the velocity by 88.1%, 85.9 and 89.6% respectively. Furthermore, for the inner rectangular split baffle angles of 0°,45°, and 90°, and within the Rayleigh number range of 104 to 106, the Nuavg exhibits significant variations, with maximum increases of 71.9% , 46.7% and a subsequent decrease of 45.9%.

On the Generalization Capability of a Data-Driven Turbulence Model by Field Inversion and Machine Learning

This paper discusses the generalizability of a data-augmented turbulence model with a focus on the field inversion and machine learning approach. It is highlighted that the augmented model based on two-dimensional (2D) separated airfoil flows gives poor predictive capability for a different class of separated flows (NASA wall-mounted hump) compared to the baseline model due to extrapolation. We demonstrate a sensor-based approach to localize the data-driven model correction to tackle this generalizability issue. Furthermore, the applicability of the augmented model to a more complex aeronautical three-dimensional case, the NASA Common Research Model configuration, is studied. Observations on the pressure coefficient predictions and the model correction field suggest that the present 2D-based augmentation is to some extent applicable to a three-dimensional aircraft flow.

Experimental Tests and Numerical Analysis of Aerodynamic Properties of the Composite-Made Jet-Propelled Aerial Target.

The design of an aircraft's internal structure, and therefore the appropriate choice of material type, is a direct function of the performed tasks and the magnitude and type of the acting loads. The design of a durable aircraft structure with appropriate stiffness and lightness requires knowledge of the loads that will be applied to the structure. Therefore, this paper presents the results of an aerodynamic experimental test and numerical analysis of a newly designed jet-propelled aerial target. The experimental tests were carried out in a low-speed wind tunnel for a wide range of angles of attack and sideslips. Moreover, they were performed for various configurations of the airplane model. In addition, the results of the experimental test were supplemented with the results of the numerical analysis performed using computational fluid dynamics methods. During numerical analysis, specialized software based on solving partial differential equations using the Finite Volumes Method was used. This article presents the methodology of the conducted research. The results of the aerodynamic analysis are presented in the form of diagrams showing the aerodynamic force and moment components as a function of the angle of attack and sideslip. In addition, qualitative results of the flow around the plane have been presented. The results obtained prove that the adopted methods are sufficient to solve these types of problem. The aerial system was positively verified during the qualification tests of the system at the Polish Air Force training range and finally received the type certificate.

From nutrients to fish: Impacts of mesoscale processes in a global CESM-FEISTY eddying ocean model framework

The ocean sustains ecosystems that are essential for human livelihood and habitability of the planet. The ocean holds an enormous amount of carbon, and serves as a critical source of nutrition for human societies worldwide. Climate variability and change impact marine biogeochemistry and ecosystems. Thus, having state-of-the-art simulations of the ocean, which include marine biogeochemistry and ecosystems, is critical for understanding the role of climate variability and change on the ocean biosphere. Here we present a novel global eddy-resolving (0.1° horizontal resolution) simulation of the ocean and sea ice, including ocean biogeochemistry, performed with the Community Earth System Model (CESM). The simulation is forced by the atmospheric dataset based on the Japanese Reanalysis (JRA-55) product over the 1958–2021 period. We present a novel configuration of the CESM marine ecosystem model in this simulation which includes two zooplankton classes: microzooplankton and mesozooplankton. This novel planktonic food web structure facilitates “offline” coupling with the Fisheries Size and Functional Type (FEISTY) model. FEISTY is a size- and trait-based model of fish functional types contributing to fisheries. We present an evaluation of the ocean biogeochemistry, marine ecosystem (including fish types), and sea ice in this high resolution simulation compared to available observations and a corresponding low resolution (nominal 1°) simulation. Our analysis offers insights into environmental controls on trophodynamics within the ocean. We find that this high resolution simulation provides a realistic reconstruction of nutrients, oxygen, sea ice, plankton and fish distributions over the global ocean. On global and large regional scales the high resolution simulation is comparable to the standard 1° simulation, but on smaller scales, explicitly resolving the mesoscale dynamics is shown to be important for accurately capturing trophodynamic structuring, especially in coastal ecosystems. We show that fine-scale ocean features leave imprints on ocean ecosystems, from plankton to fish, from the tropics to polar regions. This simulation also offers insights on ocean acidification over the past 64 years, as well as how large-scale climate variations may impact upper trophic levels. The data generated by the simulations are publicly available and will be a fruitful community resource for a large variety of oceanographic science questions.

The Role of Double Excitations in Exciton Dynamics of Multiazobenzenes: Trisazobenzenophane as a Test Case.

Molecular exciton dynamics underlie energy and charge transfer processes in organic multichromophoric systems. A particularly interesting class of the latter is multiphotochromic systems made of molecules capable of photochemical transformations. Exciton dynamics in assemblies of photoswitches have been recently investigated using either the molecular exciton model or supermolecular configuration interaction (CI) singles, both approaches being based on a semiempirical Hamiltonian and combined with surface hopping molecular dynamics. Here, we study how inclusion of double excitations in nonadiabatic dynamics simulations affects exciton dynamics of multiazobenzenes, using trisazobenzenophane as an example. We find that both CI singles and CI singles and doubles yield virtually the same time scale of dynamical exciton localization, ∼50 fs for the studied multiazobenzene. However, inclusion of double excitations considerably affects the excited state lifetimes and isomerization quantum yields.

The Met Office Forecast Ocean Assimilation Model (FOAM) using a 1/12‐degree grid for global forecasts

AbstractThe Met Office Forecast Ocean Assimilation Model (FOAM) ocean–sea‐ice analysis and forecasting operational system has been using an ORCA tripolar grid with 1/4° horizontal grid spacing since December 2008. Surface boundary forcing is provided by numerical weather prediction fields from the operational global atmosphere Met Office Unified Model. We present results from a 2‐year simulation using a 1/12° global ocean–sea‐ice model configuration while keeping a 1/4° data assimilation (DA) set‐up. We also describe recent operational data assimilation enhancements that are included in our 1/4° control and 1/12° simulations: a new bias‐correction term for sea‐level anomaly assimilation and a revised pressure correction algorithm. The primary effect of the first is to decrease the mean and variability of sea‐level anomaly increments at high latitudes, whereas the second significantly reduces the vertical velocity standard deviation in the tropical Pacific. The level of improvement achieved with the higher resolution configuration is moderate but consistently satisfactory when measured using neighbourhood verification metrics that provide fairer quantitative comparisons between gridded model fields at different spatial resolutions than traditional root‐mean‐square metrics. A comparison of the eddy kinetic energy from each configuration and an observation‐based product highlights the regions where further system developments are most needed.

Oral mucosal lesions triage via YOLOv7 models

Background/purposeThe global incidence of lip and oral cavity cancer continues to rise, necessitating improved early detection methods. This study leverages the capabilities of computer vision and deep learning to enhance the early detection and classification of oral mucosal lesions. MethodsA dataset initially consisting of 6903 white-light macroscopic images collected from 2006 to 2013 was expanded to over 50,000 images to train the YOLOv7 deep learning model. Lesions were categorized into three referral grades: benign (green), potentially malignant (yellow), and malignant (red), facilitating efficient triage. ResultsThe YOLOv7 models, particularly the YOLOv7-E6, demonstrated high precision and recall across all lesion categories. The YOLOv7-D6 model excelled at identifying malignant lesions with notable precision, recall, and F1 scores. Enhancements, including the integration of coordinate attention in the YOLOv7-D6-CA model, significantly improved the accuracy of lesion classification. ConclusionThe study underscores the robust comparison of various YOLOv7 model configurations in the classification to triage oral lesions. The overall results highlight the potential of deep learning models to contribute to the early detection of oral cancers, offering valuable tools for both clinical settings and remote screening applications.

Parameter estimation in a whole-brain network model of epilepsy: Comparison of parallel global optimization solvers.

The Virtual Epileptic Patient (VEP) refers to a computer-based representation of a patient with epilepsy that combines personalized anatomical data with dynamical models of abnormal brain activities. It is capable of generating spatio-temporal seizure patterns that resemble those recorded with invasive methods such as stereoelectro EEG data, allowing for the evaluation of clinical hypotheses before planning surgery. This study highlights the effectiveness of calibrating VEP models using a global optimization approach. The approach utilizes SaCeSS, a cooperative metaheuristic algorithm capable of parallel computation, to yield high-quality solutions without requiring excessive computational time. Through extensive benchmarking on synthetic data, our proposal successfully solved a set of different configurations of VEP models, demonstrating better scalability and superior performance against other parallel solvers. These results were further enhanced using a Bayesian optimization framework for hyperparameter tuning, with significant gains in terms of both accuracy and computational cost. Additionally, we added a scalable uncertainty quantification phase after model calibration, and used it to assess the variability in estimated parameters across different problems. Overall, this study has the potential to improve the estimation of pathological brain areas in drug-resistant epilepsy, thereby to inform the clinical decision-making process.

Hyperparameter optimization of two-branch neural networks in multi-target prediction

As a result of the ever increasing complexity of configuring and fine-tuning machine learning models, the field of automated machine learning (AutoML) has emerged over the past decade. However, software implementations like Auto-WEKA and Auto-sklearn typically focus on classical machine learning (ML) tasks such as classification and regression. Our work can be seen as the first attempt at offering a single AutoML framework for most problem settings that fall under the umbrella of multi-target prediction, which includes popular ML settings such as multi-label classification, multivariate regression, multi-task learning, dyadic prediction, matrix completion, and zero-shot learning. Automated problem selection and model configuration are achieved by extending DeepMTP, a general deep learning framework for MTP problem settings, with popular hyperparameter optimization (HPO) methods. Our extensive benchmarking across different datasets and MTP problem settings identifies cases where specific HPO methods outperform others.

Network properties determine neural network performance

Machine learning influences numerous aspects of modern society, empowers new technologies, from Alphago to ChatGPT, and increasingly materializes in consumer products such as smartphones and self-driving cars. Despite the vital role and broad applications of artificial neural networks, we lack systematic approaches, such as network science, to understand their underlying mechanism. The difficulty is rooted in many possible model configurations, each with different hyper-parameters and weighted architectures determined by noisy data. We bridge the gap by developing a mathematical framework that maps the neural network’s performance to the network characters of the line graph governed by the edge dynamics of stochastic gradient descent differential equations. This framework enables us to derive a neural capacitance metric to universally capture a model’s generalization capability on a downstream task and predict model performance using only early training results. The numerical results on 17 pre-trained ImageNet models across five benchmark datasets and one NAS benchmark indicate that our neural capacitance metric is a powerful indicator for model selection based only on early training results and is more efficient than state-of-the-art methods.

You only compress once: Towards effective and elastic BERT compression via exploit–explore stochastic nature gradient

Despite superior performance on various natural language processing tasks, pre-trained models such as BERT are challenged by deploying on resource-constraint devices. Most existing model compression approaches require re-compression or fine-tuning across diverse constraints to accommodate various hardware deployments. This practically limits the further application of model compression. Moreover, the ineffective training and searching process of existing elastic compression paradigms (Wang et al., 2020; Cai et al., 2020) prevents the direct migration to BERT compression. Motivated by the necessity of efficient inference across various constraints on BERT, we propose a novel approach, YOCO-BERT, to achieve compress once and deploy everywhere. Specifically, we first construct a huge search space with 1013 architectures, which covers nearly all configurations in BERT model. Then, we propose a novel stochastic nature gradient optimization method to guide the generation of optimal candidate architecture which could keep a balanced trade-off between explorations and exploitation. When a certain resource constraint is given, a lightweight distribution optimization approach is utilized to obtain the optimal network for target deployment without fine-tuning. Compared with state-of-the-art algorithms, YOCO-BERT provides more compact models, yet achieving 2.1%–4.5% average accuracy improvement on the GLUE benchmark. Besides, YOCO-BERT is also more effective, e.g., the training complexity is O(1) for N different devices. Codes available https://github.com/MAC-AutoML/YOCO-BERT.

Glove-Net: Enhancing Grasp Classification with Multisensory Data and Deep Learning Approach.

Grasp classification is pivotal for understanding human interactions with objects, with wide-ranging applications in robotics, prosthetics, and rehabilitation. This study introduces a novel methodology utilizing a multisensory data glove to capture intricate grasp dynamics, including finger posture bending angles and fingertip forces. Our dataset comprises data collected from 10 participants engaging in grasp trials with 24 objects using the YCB object set. We evaluate classification performance under three scenarios: utilizing grasp posture alone, utilizing grasp force alone, and combining both modalities. We propose Glove-Net, a hybrid CNN-BiLSTM architecture for classifying grasp patterns within our dataset, aiming to harness the unique advantages offered by both CNNs and BiLSTM networks. This model seamlessly integrates CNNs' spatial feature extraction capabilities with the temporal sequence learning strengths inherent in BiLSTM networks, effectively addressing the intricate dependencies present within our grasping data. Our study includes findings from an extensive ablation study aimed at optimizing model configurations and hyperparameters. We quantify and compare the classification accuracy across these scenarios: CNN achieved 88.09%, 69.38%, and 93.51% testing accuracies for posture-only, force-only, and combined data, respectively. LSTM exhibited accuracies of 86.02%, 70.52%, and 92.19% for the same scenarios. Notably, the hybrid CNN-BiLSTM proposed model demonstrated superior performance with accuracies of 90.83%, 73.12%, and 98.75% across the respective scenarios. Through rigorous numerical experimentation, our results underscore the significance of multimodal grasp classification and highlight the efficacy of the proposed hybrid Glove-Net architectures in leveraging multisensory data for precise grasp recognition. These insights advance understanding of human-machine interaction and hold promise for diverse real-world applications.

Early detection of pediatric health risks using maternal and child health data

Machine learning (ML)-driven diagnosis systems are particularly relevant in pediatrics given the well-documented impact of early-life health conditions on later-life outcomes. Yet, early identification of diseases and their subsequent impact on length of hospital stay for this age group has so far remained uncharacterized, likely because access to relevant health data is severely limited. Thanks to a confidential data use agreement with the California Department of Health Care Access and Information, we introduce Ped-BERT: a state-of-the-art deep learning model that accurately predicts the likelihood of 100+ conditions and the length of stay in a pediatric patient’s next medical visit. We link mother-specific pre- and postnatal period health information to pediatric patient hospital discharge and emergency room visits. Our data set comprises 513.9K mother–baby pairs and contains medical diagnosis codes, length of stay, as well as temporal and spatial pediatric patient characteristics, such as age and residency zip code at the time of visit. Following the popular bidirectional encoder representations from the transformers (BERT) approach, we pre-train Ped-BERT via the masked language modeling objective to learn embedding features for the diagnosis codes contained in our data. We then continue to fine-tune our model to accurately predict primary diagnosis outcomes and length of stay for a pediatric patient’s next visit, given the history of previous visits and, optionally, the mother’s pre- and postnatal health information. We find that Ped-BERT generally outperforms contemporary and state-of-the-art classifiers when trained with minimum features. We also find that incorporating mother health attributes leads to significant improvements in model performance overall and across all patient subgroups in our data. Our most successful Ped-BERT model configuration achieves an area under the receiver operator curve (ROC AUC) of 0.927 and an average precision score (APS) of 0.408 for the diagnosis prediction task, and a ROC AUC of 0.855 and APS of 0.815 for the length of hospital stay task. Further, we examine Ped-BERT’s fairness by determining whether prediction errors are evenly distributed across various subgroups of mother–baby demographics and health characteristics, or if certain subgroups exhibit a higher susceptibility to prediction errors.

Determinant evaluations inspired by Di Francesco's determinant for twenty-vertex configurations

In his work on the twenty vertex model, Di Francesco (2021) found a determinant formula for the number of configurations in a specific such model, and he conjectured a closed form product formula for the evaluation of this determinant. We prove this conjecture here. Moreover, we actually generalize this determinant evaluation to a one-parameter family of determinant evaluations, and we present many more determinant evaluations of similar type — some proved, some left open as conjectures.

Enhancing open stope stability prediction in mining engineering: Optimal configuration of an artificial neural network model

Ensuring the stability of underground excavations is a crucial concern in mining engineering, directly impacting operational safety and efficiency. Given the dynamic nature of the subsurface environment, innovative methods are essential to enhance stability in open stopes and mitigate potential risks. This study focuses on refining predictions of underground excavation stability in mining engineering through the optimization of an artificial neural network (ANN) model. Analyzing Potvin’s database, which includes 175 historical cases, we examined the impact of various ANN model configurations. Our findings indicate that normalizing data using Standard Scaler and employing Swish as the activation function across all layers yielded the most accurate predictions for this specific scenario. Additionally, employing the SHAP (Shapley Additive exPlanations) tool enabled us to assess feature importance and identify the most influential factors. Our results underscore the significant impact of the shape factor on underground opening stability, closely followed by the Q value. This research contributes to enhancing predictive models for underground mining excavation stability, emphasizing the critical role of specific parameters affecting open stope stability.

Advanced Lung Disease Detection and Classification Using Ge-U-Net-ODLwith Gabor Filters and Entropy-Based Feature Selection

Lung X-ray images play a crucial role in the detection and diagnosis of various lung diseases,including pneumonia, tuberculosis, and lung cancer. These images provide a non-invasive method for visualizing lung structures, allowing radiologists and machine learning models to identify abnormalities such as nodules, masses, or fluid accumulation. With the advancement of deep learning techniques, lung X-ray images are now used in automated systems that can detect and classify diseases with high accuracy. By applying sophisticated algorithms like GE-U-Net-ODL, Gabor filters, and entropy-based feature extraction, these images are analyzed pixel by pixel to enhance feature representation and improve diagnostic precision. This paper presents a novel approach for lung disease detection and classification using the GE-U-Net-ODL model, which integrates advanced preprocessing techniques and deep learning architectures. The study leverages the NIH Chest X-ray dataset and employs a variety of feature extraction and selection methods, including Gabor filters, entropy-based techniques, and multi-scale inputs. A detailed comparative analysis of different model configurations demonstrates that the GE-U-Net-ODL model with Transfer Learning achieves the highest classification accuracy of 95.0%, alongside superior precision (93.5%), recall (94.0%), and F1-score (93.7%). Other configurations, such as those utilizing data augmentation and hybrid filters, also showed notable performance improvements. The research underscores the model's effectiveness in enhancing diagnostic accuracy for lung diseases while balancing training and inference times.

Seismic Response Simulation of a Reinforced Concrete Wall-Frame Building That Collapsed During the February 2023 Kahramanmaras Earthquake

ABSTRACT In this study, the seismic response of a newly constructed RC building, which suffered collapse-level damage during the February 6, 2023 Kahramanmaras earthquake, is simulated through nonlinear response history analyses, using model configurations representing the as-built state of the building, as well as hypothetical code-compliant and CFRP-strengthened states. Analysis results show that in its as-built state, the collapse of the structure was not unexpected, whereas code-compliant confinement detailing would have probably resulted in collapse-prevention-level performance, and CFRP-strengthening of the building would have significantly improved its seismic performance and would likely have prevented total collapse.

Combination Method Of Level Shift Pulse Width Modulation With Carrier Frequency Variables On A Bridge Model Inverter

Switching is very important in energy conversion systems. The conversion process requires a good and reliable method. This research identifies methods or techniques for pulse width modulation. This technique proposes a new model of modulation technique, namely by combining level shifted modulation techniques, and adjusting the frequency in such a way that it produces a signal that suits the inverter's needs. The inverter tested in this research is an inverter with a bridge model configuration. The proposed research and methods were verified using MATLAB and Simulink software. The results obtained are the ratio of total harmonic distortion (THD) are 25.07%, 76, 66% and 9.183%, respectively

Enhancement of Quantum Efficiency in Perovskite Solar Cells Through Whispering Gallery Modes from Titanium Oxide Micro‐Resonators

With the aid of 3D full‐field finite difference time–domain simulations, model configurations for thin‐film solar cell devices that include periodically arranged microspheres, exhibiting resonating whispering gallery modes (WGMs), are proposed. The microspheres present, either immersed in perovskite or coated with perovskite layer, between the electron‐ and hole‐transport layers show enhanced current‐conversion efficiency. The presence of WGMs lead to enhancement in the absorption of layer. The incoming electromagnetic wave couples with microsphere and forms confined resonating modes. Different designs are examined for deciding the appropriate position of WGM exhibiting spheres with respect to thin‐film perovskite solar cell (PSC) featuring back reflector and optimized antireflectance coating. Since the incoupling element is lossless, energy stored in microspheres is absorbed efficiently by the underlying active material. This directly contributes to the increment in the current density of the solar cell. Thus, the devices show a higher current density of 23.62 mA cm−1, while that in planar solar cell device shows current density of 13.68 mA cm−1, for the same thickness of perovskite layer. This leads to more than 70% enhancement in the short‐circuit current density than the conventional PSCs device of similar size.

AN INFORMATION-THEORETIC FILTER SINGLE-STEP SEQUENTIAL MONTE CARLO METHOD IN PRACTICE. AN EXPERIMENTAL CASE STUDY FOR THE PARTICLE SIZE DISTRIBUTION ESTIMATION

In this article, we implement an extended version of a Monte Carlo methodology making use of a Sampling/Mapping/Filtering (SMF) strategy, to estimate the Particle Size Distribution (PSD) of a particle system. The methodological difference is shown by the use of a filter, based on information theory, which computes a cut-off level using an optimization scheme following the principle of relevant information (PRI). The practical application of the full strategy considers the fusion between scanning electron microscope (SEM) data and static light scattering (SLS) measurements. The two models are utilized, assuming spherical particles represented as hard spheres: the so-called local monodisperse approximation (LMA) and the Vrij’s finite mixtures model (VFMM). We analyze the resulting estimates for different configurations of both prior information and SLS models, where final results are compared to those obtained in previous studies for a polymeric particle system embedded in a solid polymer matrix.