Ability Of Model Research Articles

Innovative modeling on the effects of low-temperature stress on rice yields

Abstract The increasing frequency and intensity of low-temperature events in temperate and cold rice production regions threaten rice yields under climate change. While process-based crop models can project climate impacts on rice yield, their accuracy under low-temperature conditions has not been well-evaluated. Our six-year chamber experiments revealed that low temperatures reduce spikelet fertility from panicle initiation to flowering, grain number per spike during panicle development, and grain weight during grain filling. We examined the algorithms of spikelet fertility response to temperature used in crop models. Results showed that simulation performance is poor for crop yields if the same function was used at different growth stages outside the booting stage. Indeed, we replaced a parameter spikelet fertility algorithm of the ORYZA model and developed the function of estimating grain number per spike and grain weight. After that, the improved equation algorithm was applied to 10 rice growth models. New functions considered the harmful effects of low temperatures on rice yield at different stages. In addition, the threshold temperatures of the cold tolerance were set for different rice varieties. The improved algorithm enhances the model's ability to simulate rice yields under climate change, providing a more reliable tool for adapting rice production to future climatic challenges.

Deep Learning-Based Postoperative Glioblastoma Segmentation and Extent of Resection Evaluation: Development, External Validation, and Model Comparison

Abstract Background The pursuit of automated methods to assess the extent of resection (EOR) in glioblastomas is challenging, requiring precise measurement of residual tumor volume. Many algorithms focus on preoperative scans, making them unsuitable for postoperative studies. Our objective was to develop a deep learning-based model for postoperative segmentation using magnetic resonance imaging (MRI). We also compared our model's performance with other available algorithms. Methods To develop the segmentation model, a training cohort from three research institutions and three public databases was used. Multiparametric MRI scans with ground truth labels for contrast enhancing tumor, edema, and surgical cavity, served as training data. The models were trained using MONAI and nnU-Net frameworks. Comparisons were made with currently available segmentation models using an external cohort from a research institution and a public database. Additionally, the model's ability to classify EOR was evaluated using the RANO-Resect classification system. To further validate our best-trained model, an additional independent cohort was used. Results The study included 586 scans: 395 for model training, 52 for model comparison, and 139 scans for independent validation. The nnU-Net framework produced the best model with median Dice scores of 0.81 for contrast enhancing tumor, 0.77 for edema, and 0.81 for surgical cavity. Our best-trained model classified patients into maximal and submaximal resection categories with 96% accuracy in the model comparison dataset and 84% in the independent validation cohort. Conclusion Our nnU-Net-based model outperformed other algorithms in both segmentation and EOR classification tasks, providing a freely accessible tool with promising clinical applicability.

Risk factors and discrimination model for screen exposure among children under 3 years: findings from Fujian province in China.

Compliance with screen time guidelines among children worldwide remains low, and there is insufficient evidence on the current prevalence in China. This study aimed to investigate the prevalence of compliance with screen time guidelines among children under 3 years old in Fujian Province, East China, identify risk factors and their independent effects, and develop a risk discrimination model for targeted interventions. A cross-sectional survey was conducted among low-income families recruited from welfare programs at 96 sites in both urban and rural areas of Fujian Province, China. Face-to-face interviews gathered sociodemographic data, lifestyle information, attitudes towards screen exposure, and details on screen media device ownership. A multivariable logistic regression model was employed to identify independent risk factors for compliance with screen time guidelines, while the area under the receiver operating characteristic curve (AUC) was used to evaluate the model's discrimination ability. A total of 4,707 children participated in the survey. The rates of compliance with screen time guidelines were 56.8% for children under 1 year old, 18.8% for those between 1 and 2 years old, and 81.9% for those between 2 and 3 years old. The multivariable regression analysis identified negative attitudes towards screen exposure, co-viewing and engagement, as well as single child, as significant positive independent factors for compliance with the guidelines. The risk discrimination model demonstrated good performance, with an AUC of 0.845 and 0.812 in the two younger age groups, but showed medium discrimination with an AUC of 0.691 for children between 2 and 3 years old. Compliance with screen time guidelines among young children in Fujian Province, East China, is generally adequate, but notably low among children between 1 and 2 years old. Targeted interventions are needed to improve compliance, particularly for this age group.

Learning spatial‐frequency interaction for generalizable deepfake detection

AbstractIn recent years, face forgery detection has gained significant attention, resulting in considerable advancements. However, most existing methods rely on CNNs to extract artefacts from the spatial domain, overlooking the pervasive frequency‐domain artefacts present in deepfake content, which poses challenges in achieving robust and generalized detection. To address these issues, we propose the dual‐stream frequency—spatial fusion network is proposed for deepfake detection. The dual‐stream frequency‐spatial fusion network consists of three components: the spatial forgery feature extraction module, the frequency forgery feature extraction module, and the spatial–frequency feature fusion module. The spatial forgery feature extraction module employs spatial‐channel attention to extract spatial domain features, targeting artefacts in the spatial domain. The frequency forgery feature extraction module leverages the focused linear attention to detect frequency domain anomalies in internal regions, enabling the identification of generated content. The spatial–frequency feature fusion module then fuses forgery features extracted from both the spatial and frequency domains, facilitating accurate detection of splicing artefacts and internally generated forgeries. This approach enhances the model's ability to more accurately capture forgery characteristics. Extensive experiments on several widely‐used benchmarks demonstrate that our carefully designed network exhibits superior generalization and robustness, significantly improving deepfake detection performance.

Physics-guided spatiotemporal approach for melt pool size prediction in laser powder bed fusion

ABSTRACT This study first proposes a physics-guided spatiotemporal graph. The melt pool temporal feature nodes and spatial feature nodes are extracted from melt pool images based on spatiotemporal correlations. These feature points are treated as graph nodes, and the edge weights in the graph are calculated based on an analytical heat transfer model. It supplements the physical information of accumulated heat. Based on this, the graph convolutional network (GCN) is used to extract spatiotemporal information of heat transfer, and the long short-term memory (LSTM) is used to capture the evolution patterns of the melt pool. Thus, a melt pool spatiotemporal feature extraction model is constructed for melt pool size prediction. The model achieves prediction errors within 12% for the melt pool area and 8.5% for the melt pool perimeter. Subsequently, interpretability analysis indicates that the physics-guided approach enhances the model's learning ability for the cumulative contribution of heat from neighboring tracks.

Multi-kernel inception aggregation diffusion network for tomato disease detection.

Tomato leaf diseases significantly impact the yield and quality of tomatoes during cultivation, the main of which are septoria leaf spot, leaf curl virus, verticillium wilt, and early blight. These diseases necessitate prompt detection and management strategies to mitigate their deleterious effects on crop productivity. Due to the considerable scale variations in diseased tomato leaves, accurate and rapid detection and diagnosis remain challenging. To address the detection of tomato leaf diseases at different scales, we propose a real-time detection model incorporating a Multi-kernel Inception Aggregation Diffusion Network. In this paper, (1) We present a Multi-kernel Inception Aggregation Diffusion Network (MIADN) for the feature processing stage, which facilitates the aggregation and diffusion of multi-scale features across hierarchical levels, benefiting the detection of targets at various scales. (2) We present the Multi-kernel Inception Module (MKIM), designed to extract multi-scale object features using diverse convolution kernels, thereby enhancing the model's feature fusion and representation capabilities. (3) We incorporate the efficient FasterNet network at the feature extraction stage to preserve feature diversity and improve the model's ability to extract complex target features. (4) Extensive comparative and ablation experiments demonstrate that our method achieves the mean average precision (mAP50) of 96.6%, surpassing the baseline model by 4.1% and the advanced YOLOv9s model by 2.0%. This method provides an effective solution for high-quality tomato cultivation.

Development and validation of a risk prediction model for severe postoperative complications in elderly patients with hip fracture.

This study aimed to investigate risk factors associated with severe postoperative complications following hip fracture surgery in elderly patients and to develop a nomogram-based risk prediction model for these complications. A total of 627 elderly patients with hip fractures treated at Yongchuan Hospital of Chongqing Medical University from January 2015 to April 2024 were collected. 439 patients were assigned to the training cohort for model development, and 188 to the validation cohort for model assessment. The training cohort was stratified based on the presence or absence of severe complications. We employed LASSO regression, as well as univariate and multivariate logistic regression analyses, to identify significant factors. A nomogram was constructed based on the outcomes of the multivariate regression. The model's discriminative ability was assessed using the area under the receiver operating characteristic curve (AUC), while calibration plots and decision curve analysis (DCA) evaluated its calibration and stability. Internal validation was performed using the validation cohort. Out of the 627 patients, 118 (18.82%) experienced severe postoperative complications. Both LASSO regression and multivariate logistic analysis identified the modified 5-item frailty index (mFI-5) and the preoperative C-reactive protein to albumin ratio (CAR) as significant predictors of severe complications. The nomogram model, derived from the multivariate analysis, exhibited strong discriminative ability, with an AUC of 0.963 (95% CI: 0.946-0.980) for the training cohort and 0.963 (95% CI: 0.938-0.988) for the validation cohort. Calibration plots demonstrated excellent agreement between the nomogram's predictions and actual outcomes. Decision curve analysis (DCA) indicated that the model provided clinical utility across all patient scenarios. These findings were consistent in the validation cohort. Both the mFI-5 and CAR are predictive factors for severe postoperative complications in elderly patients undergoing hip fracture surgery.

A novel approach for multivariate time series interval prediction of water quality at wastewater treatment plants

ABSTRACT This study proposes a novel approach for predicting variations in water quality at wastewater treatment plants (WWTPs), which is crucial for optimizing process management and pollution control. The model combines convolutional bi-directional gated recursive units (CBGRUs) with Adaptive Bandwidth Kernel Function Density Estimation (ABKDE) to address the challenge of multivariate time series interval prediction of WWTP water quality. Initially, wavelet transform (WT) was employed to smooth the water quality data, reducing noise and fluctuations. Linear correlation coefficient (CC) and non-linear mutual information (MI) techniques were then utilized to select input variables. The CBGRU model was applied to capture temporal correlations in the time series, integrating the Multiple Heads of Attention (MHA) mechanism to enhance the model's ability to comprehend complex relationships within the data. ABKDE was employed, supplemented by bootstrap to establish upper and lower bounds of the prediction intervals. Ablation experiments and comparative analyses with benchmark models confirmed the superior performance of the model in point prediction, interval prediction, the analysis of forecast period, and fluctuation detection for water quality data. Also, this study verifies the model's broad applicability and robustness to anomalous data. This study contributes significantly to improved effluent treatment efficiency and water quality control in WWTPs.

An Enhanced Detection Method of Hardware Trojan Based on CNN- Attention-LSTM

The security issues caused by the insertion of hardware Trojans seriously threaten the security and reliability of the entire hardware device. This article constructs a detection model that combines convolutional neural networks (CNN) and long short-term memory networks (LSTM), and introduces attention mechanism to enhance the model's ability to recognize complex circuits. This method can automatically learn and optimize feature extraction and classification models, reduce reliance on manual experience through training on large amounts of data, and improve the intelligence level of detection. Especially, by combining attention mechanisms and LSTM models, it is possible to more effectively capture small anomalies in circuit design and improve the accuracy and efficiency of hardware Trojan detection. The experimental results show that the proposed CNN-Attention-LSTM model exhibits superior Trojan detection performance and good generalization ability on different datasets, with an precision of 96.3%, a recall rate of 94.7%, and an F1 score of 95.5%.

Machine learning-based prediction of bioactivity in HIV-1 protease: insights from electron density analysis.

Aim: To develop a model for predicting the biological activity of compounds targeting the HIV-1 protease and to establish factors influencing enzyme inhibition.Materials & methods: Machine learning models were built based on a combination of Richard Bader's theory of Atoms in Molecules and topological analysis of electron density using experimental x-ray 'protein-ligand' complexes and inhibition constants data.Results & conclusion: Among all the models tested, logistic regression achieved the highest accuracy of 0.76 on the test set. The model's ability to differentiate between less active and highly active classes was relatively good, as indicated by an AUC-ROC score of 0.77. The analysis identified several critical factors affecting the biological activity of HIV-1 protease inhibitors, including the electron density contribution of hydrogen atoms, bond-critical points and particular amino acid residues. These findings provide new insights into how these molecular factors influence HIV-1 protease inhibition, emphasizing the importance of hydrogen bonding, glycine's flexibility and hydrophobic interactions in ligand binding.

ECG-based machine learning model for AF identification in patients with first ischemic stroke.

The recurrence rate of strokes associated with atrial fibrillation (AF) can be substantially reduced through the administration of oral anticoagulants. However, previous studies have not demonstrated a clear benefit from the universal application of oral anticoagulants in patients with embolic stroke of undetermined source. Timely detection of AF remains a challenge in patients with stroke. This study aims to develop a convolutional neural network (CNN) model to accurately identify patients with AF using a 12-lead sinus-rhythm electrocardiogram (ECG) recorded around the time of the first ischemic stroke. Additionally, this study also evaluates the model's ability to predict future occurrence of AF. A CNN model was trained with ECG data from patients at Taipei Veterans General Hospital. External validation was performed on ischemic stroke patients from National Taiwan University Hospital. The model's performance was assessed for detecting AF at the stroke event and predicting future AF occurrences. The model demonstrated an area under curve (AUC) of 0.91 for internal validation and 0.69 for external validation in identifying AF at the stroke event, with sensitivity and negative predictive value both achieving 97%. Kaplan-Meier survival analysis of patients without a prior diagnosis of AF revealed a significant increase in future AF incidence among the high-risk group identified by the model (adjusted hazard ratio: 4.06; 95% confidence interval: 2.74-6.00). The CNN model effectively identifies AF in stroke patients using 12-lead ECGs and predicts future AF events, facilitating early anticoagulation therapy and potentially reducing recurrent stroke risk. Further prospective studies are warranted to confirm these findings.

Enhancing image-based diagnosis of gastrointestinal tract diseases through deep learning with EfficientNet and advanced data augmentation techniques.

The early detection and diagnosis of gastrointestinal tract diseases, such as ulcerative colitis, polyps, and esophagitis, are crucial for timely treatment. Traditional imaging techniques often rely on manual interpretation, which is subject to variability and may lack precision. Current methodologies leverage conventional deep learning models that, while effective to an extent, often suffer from overfitting and generalization issues on medical image datasets due to the intricate and subtle variations in disease manifestations. These models typically do not fully utilize the potential of transfer learning or advanced data augmentation, leading to less-than-optimal performance, especially in diverse real-world scenarios where data variability is high. This study introduces a robust model using the EfficientNetB5 architecture combined with a sophisticated data augmentation strategy. The model is tailored for the high variability and intricate details present in gastrointestinal tract disease images. By integrating transfer learning with maximal pooling and extensive regularization, the model aims to enhance diagnostic accuracy and reduce overfitting. The proposed model achieved a test accuracy of 98.89%, surpassing traditional methods by incorporating advanced regularization and augmentation techniques. The application of horizontal flipping and dynamic scaling during training significantly improved the model's ability to generalize, evidenced by a low-test loss of 0.230 and high precision metrics across all classes. The proposed deep learning framework demonstrates superior performance in the automated classification of gastrointestinal diseases from image data. By addressing key limitations of existing models through innovative techniques, this study contributes to the enhancement of diagnostic processes in medical imaging, potentially leading to more accurate and timely disease interventions.

Abstract 4134729: Quantitative, time-efficient viability cardiac magnetic resonance with delayed-phase dynamic contrast enhancement model: a pilot study in dogs with myocardial infarction

Introduction: Late gadolinium enhancement (LGE) is the gold standard for myocardial viability imaging, recommended by all major cardiology societies. In this study, we aimed to develop a time-efficient viability imaging technique utilizing a dynamic contrast enhancement(dDCE) model to shorten the LGE wait time, provide quantitative measures of the contrast washout procedure, and potentially enable a fast and quantitative mean for scar imaging. Methods: A canine study(n=10) was conducted on reperfused myocardial infarction(MI). A dDCE model using dynamic post-contrast T1 maps was adopted to depict the contrast washout process. dDCE maps were reconstructed using the whole dataset(dDCE 30min ) and a 5-minute subset of the T1 maps(dDCE 5min ). To test the dDCE models for shortening the LGE wait time, LGE dDCE images were synthesized using the dDCE 5min maps and compared to the clinical LGE images. Remote and MI dDCE parameters on dDCE 30min and dDCE 5min maps were compared to test the model's ability to extract physiological features. Results: dDCE 30min and dDCE 5min map showed a good visuospatial difference between remote and MI(Fig. 1A) and no quantitative difference (Fig.1B). v e and PS showed a significant elevation in MI than in remote (61.12±13.65% vs 13.43±5.00%, P =0.018; 44.62±21.40mL/g/min vs 0.42±0.55mL/g/min, P =0.018, respectively). v p and F p were significantly decreased in MI than in remote (4.17±2.06% vs 10.85±3.77%, P =0.02; 1.11±0.32mL/g/min vs 1.51±0.42mL/g/min, P =0.04, respectively). Representative LGE and LGE dDCE images showed high agreement in the presence of MI (Fig. 2A). Bland-Altman analysis showed good agreement between LGE and LGE dDCE images for the measurement of infarct area (bias, -1.74 ± 6.60 %, Fig. 2B) and transmuraltiy (bias, 1.86 ± 2.73 %, Fig. 2C). The simple liner regression showed strong correlations between LGE and LGE dDCE images for the infarct area (R 2 , 0.95; slope, 0.93, P <0.01; Fig. 2D) and transmurality (R 2 , 0.97; slope, 0.93, P <0.01; Fig. 2E). ROC analysis showed that the AUC was 0.97 (95% CI: 0.94 to 1.00) (Fig. 2F). Increased lesion-blood pool contrast by 10 folds on dDCE images improved subendocardial MI detection (Fig.3). Conclusions: The developed dDCE model provides comparable viability assessment ability to the standard LGE images without the prolonged wait time and reflects MI pathophysiological changes. It shows the potential for a rapid and quantitative myocardial viability imaging protocol using cardiac magnetic resonance.

Enhancing multiclass COVID-19 prediction with ESN-MDFS: Extreme smart network using mean dropout feature selection technique.

Deep learning and artificial intelligence offer promising tools for improving the accuracy and efficiency of diagnosing various lung conditions using portable chest x-rays (CXRs). This study explores this potential by leveraging a large dataset containing over 6,000 CXR images from publicly available sources. These images encompass COVID-19 cases, normal cases, and patients with viral or bacterial pneumonia. The research proposes a novel approach called "Enhancing COVID Prediction with ESN-MDFS" that utilizes a combination of an Extreme Smart Network (ESN) and a Mean Dropout Feature Selection Technique (MDFS). This study aimed to enhance multi-class lung condition detection in portable chest X-rays by combining static texture features with dynamic deep learning features extracted from a pre-trained VGG-16 model. To optimize performance, preprocessing, data imbalance, and hyperparameter tuning were meticulously addressed. The proposed ESN-MDFS model achieved a peak accuracy of 96.18% with an AUC of 1.00 in a six-fold cross-validation. Our findings demonstrate the model's superior ability to differentiate between COVID-19, bacterial pneumonia, viral pneumonia, and normal conditions, promising significant advancements in diagnostic accuracy and efficiency.

Exploration of an intrinsically explainable self-attention based model for prototype generation on single-channel EEG sleep stage classification.

Prototype-based methods in deep learning offer interpretable explanations for decisions by comparing inputs to typical representatives in the data. This study explores the adaptation of SESM, a self-attention-based prototype method successful in electrocardiogram (ECG) tasks, for electroencephalogram (EEG) signals. The architecture is evaluated on sleep stage classification, exploring its efficacy in predicting stages with single-channel EEG. The model achieves comparable test accuracy compared to EEGNet, a state-of-the-art black-box architecture for EEG classification. The generated prototypical components are exaimed qualitatively and using the area over the perterbation curve (AOPC) indicate some alignment with expected bio-markers for different sleep stages such as alpha spindles and slow waves in non-REM sleep, but the results are severely limited by the model's ability to only extract and present information in the time-domain. Ablation studies are used to explore the impact of kernel size, number of heads, and diversity threshold on model performance and explainability. This study represents the first application of a self-attention based prototype method to EEG data and provides a step forward in explainable AI for EEG data analysis.

Baseline predictors associated with successful weight loss among breast cancer survivors in the Lifestyle, Exercise, and Nutrition (LEAN) study.

To investigate participant characteristics associated with clinically meaningful weight loss (≥ 5% weight loss) among breast cancer survivors participating in the Lifestyle, Exercise, and Nutrition (LEAN) study. Female breast cancer survivors with a body mass index ≥ 25kg/m2 were enrolled in a 6-month weight loss intervention. Univariate and multivariable logistic regression were used to determine baseline predictors of clinically meaningful weight loss (< vs. ≥ 5%) among those in the intervention arm. The area under the receiver operator characteristic curve (ROC-AUC) C-statistic evaluated the final model's ability to classify weight loss success. Baseline data were available for 77 participants. Having a history of weight gain during breast cancer treatment was associated with lower odds of achieving ≥ 5% weight loss (OR = 0.36, 95% CI 0.13 - 0.99) compared to women with no history of weight gain during cancer treatment. Breast cancer survivors with overweight or obesity who gained weight during their cancer treatment were less likely to achieve clinically meaningful weight loss during a post-treatment weight loss intervention. Breast cancer survivors are encouraged to achieve and maintain a healthy body weight to minimize the risk for cancer recurrence and comorbidity. Our results indicate that programs aimed at prevention of weight gain during treatment could aid in this goal. Understanding characteristics that increase or decrease the likelihood of achieving clinically meaningful weight loss will inform the design of programs that better support breast cancer survivors' weight loss success.

Asymmetric Large Kernel Distillation Network for efficient single image super-resolution

Recently, significant advancements have been made in the field of efficient single-image super-resolution, primarily driven by the innovative concept of information distillation. This method adeptly leverages multi-level features to facilitate high-resolution image reconstruction, allowing for enhanced detail and clarity. However, many existing approaches predominantly emphasize the enhancement of distilled features, often overlooking the critical aspect of improving the feature extraction capabilities of the distillation module itself. In this paper, we address this limitation by introducing an asymmetric large-kernel convolution design. By increasing the size of the convolution kernel, we expand the receptive field, which enables the model to more effectively capture long-range dependencies among image pixels. This enhancement significantly improves the model's perceptual ability, leading to more accurate reconstructions. To maintain a manageable level of model complexity, we adopt a lightweight architecture that employs asymmetric convolution techniques. Building on this foundation, we propose the Lightweight Asymmetric Large Kernel Distillation Network (ALKDNet). Comprehensive experiments conducted on five widely recognized benchmark datasets-Set5, Set14, BSD100, Urban100, and Manga109-indicate that ALKDNet not only preserves efficiency but also demonstrates performance enhancements relative to existing super-resolution methods. The average PSNR and SSIM values show improvements of 0.10 dB and 0.0013, respectively, thereby achieving state-of-the art performance.

Automatic segmentation and visualization of cortical and marrow bone in mandibular condyle on CBCT: a preliminary exploration of clinical application.

To develop a deep learning-based automatic segmentation method for cortex and marrow in mandibular condyle on cone-beam computed tomography (CBCT) images and explore its clinical application. 825 condyles of 490 CBCT images from 3 centers of Stomatology hospitalaffliated to Zhejiang University School of Medicinewere collected. A deep learning model was developed for simultaneous segmentation of cortex and marrow in mandibular condyle. It included a region of interest extraction network and a segmentation network based on 3D U-net, with modifications made to improve the segmentation boundaries. To evaluate its clinical potential, the model's segmentation efficiency and accuracy were compared with those of both junior and senior oral and maxillofacial radiologists. Additionally, the model's ability to assist junior radiologists in diagnosis through visualization and quantitative analysis of the generated 3D model was also assessed. The Dice similarity coefficient of the deep learning model was 0.901 (cortex), 0.969 (marrow), and 0.982 (entire condyle). Hausdorff distance was 0.755mm (cortex), 0.826mm (marrow), and 0.760mm (entire condyle). The model outperformed radiologists across all segmentation metrics, completing the task in merely 15.06s. With the assistance of visualization and quantitative analysis generated from the model's segmentation, the diagnostic accuracy of junior radiologists significantly improved. The proposed deep learning-based model achieved accurate and efficient segmentation for mandibular condylar cortex and marrow. It possessed capability to generate precise 3D models, facilitating visual quantitative measurement and aiding in the diagnosis of condylar bony changes. This model holds potential for clinical applications in orthognathic surgery, orthodontic treatment, and other TMJ-related interventions.

A Comprehensive Comparative Study of Intuitive Physics Modeling in Machine Learning Trained with Cartoon and Realistic Data

Abstract. This study delves into the influence of training data typesspecifically cartoon versus realistic visual datasetson the development of intuitive physics modeling in machine learning. Intuitive physics, the inherent human ability to understand and predict the physical properties and dynamics of objects, presents a significant challenge for current AI systems to replicate accurately. Leveraging YOLOv5, a cutting-edge object detection model, this research systematically evaluates the cognitive understanding and performance of AI models trained on distinct types of visual data. The findings reveal that the visual complexity inherent in the training datasets plays a crucial role in shaping the model's ability to generalize and accurately perform intuitive physics tasks. Models trained on cartoon datasets exhibited different learning patterns and generalization capabilities compared to those trained on realistic data, providing valuable insights into the role of data representation in AI training. This research offers both theoretical advancements in understanding AI's cognitive limitations and practical guidance for designing AI systems that can interact with the physical world more effectively. Ultimately, the study contributes to bridging the gap between human cognition and machine learning, pushing the boundaries of what AI can achieve in modeling complex, real-world phenomena.

Modelling snowpack on ice surfaces with the ORCHIDEE land surface model: application to the Greenland ice sheet

Abstract. Current climate warming is accelerating mass loss from glaciers and ice sheets. In Greenland, the rates of mass changes are now dominated by changes in surface mass balance (SMB) due to increased surface melting. To improve the future sea-level rise projections, it is therefore critical to have an accurate estimate of the SMB, which depends on the representation of the processes occurring within the snowpack. The Explicit Snow (ES) scheme implemented in the land surface model Organising Carbon and Hydrology In Dynamic Ecosystems (ORCHIDEE) has not yet been adapted to ice-covered areas. Here, we present the preliminary developments we made to apply the ES model to glaciers and ice sheets. Our analysis mainly concerns the model's ability to represent ablation-related processes. At the regional scale, our results are compared to the MAR regional atmospheric model outputs and to MODIS albedo retrievals. Using different albedo parameterizations, we performed offline ES simulations forced by the MAR model over the 2000–2019 period. Our results reveal a strong sensitivity of the modelled SMB components to the albedo parameterization. Results inferred with albedo parameters obtained using a manual tuning approach present very good agreement with the MAR outputs. Conversely, with the albedo parameterization used in the standard ORCHIDEE version, runoff and sublimation were underestimated. We also tested parameters found in a previous data assimilation experiment, calibrating the ablation processes using MODIS snow albedo. While these parameters greatly improve the modelled albedo over the entire ice sheet, they degrade the other model outputs compared to those obtained with the manually tuned approach. This is likely due to the model overfitting to the calibration albedo dataset without any constraint applied to the other processes controlling the state of the snowpack. This underlines the need to perform a “multi-objective” optimization using auxiliary observations related to internal snowpack processes. Although there is still room for further improvements, the developments reported in the present study constitute an important advance in assessing the Greenland SMB with possible extension to mountain glaciers or the Antarctic ice sheet.