Training Data Research Articles

Wildfire CNN: An Enhanced Wildfire Detection Model Leveraging CNN and VIIRS in Indian Context

Introduction Wildfires are an unexpected global hazard that significantly impact environmental change. An accurate and affordable method of identifying and monitoring on wildfire areas is to use coarse spatial resolution sensors, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS). Compared to MODIS, wildfire observations from VIIRS sensor data are around three times as extensive. Objective The traditional contextual wildfire detection method using VIIRS data mainly depends on the threshold value for classifying the fire or no fire which provides less performance for detecting wildfire areas and also fails in detecting small fires. In this paper, a wildfire detection method using Wildfiredetect Convolution Neural Network model is proposed for an effective wildfire detection and monitoring system using VIIRS data. Methods The proposed method uses the Convolutional Neural Network model and the study area dataset containing fire and non-fire spots is tested. The performance metrics such as recall rate, precision rate, omission error, commission error, F-measure and accuracy rate are considered for the model evaluation. Results The experimental analysis of the study area shows a 99.69% recall rate, 99.79% precision rate, 0.3% omission error, 0.2% commission error, 99.73% F-measure and 99.7% accuracy values for training data. The proposed method also proves to detect small fires in Alaska forest dataset for the testing data with 100% recall rate, 99.2% precision rate, 0% omission error, 0.7% commission error, 99.69% F-measure and 99.3% accuracy values. The proposed model achieves a 26.17% higher accuracy rate than the improved contextual algorithm. Conclusion The experimental findings demonstrate that the proposed model identifies small fires and works well with VIIRS data for wildfire detection and monitoring systems.

Mapping sugarcane globally at 10 m resolution using Global Ecosystem Dynamics Investigation (GEDI) and Sentinel-2

Abstract. Sugarcane is an important source of food, biofuel, and farmer income in many countries. At the same time, sugarcane is implicated in many social and environmental challenges, including water scarcity and nutrient pollution. Currently, few of the top sugar-producing countries generate reliable maps of where sugarcane is cultivated. To fill this gap, we introduce a dataset of detailed sugarcane maps for the top 13 producing countries in the world, comprising nearly 90 % of global production. Maps were generated for the 2019–2022 period by combining data from Global Ecosystem Dynamics Investigation (GEDI) and Sentinel-2 (S2). GEDI data were used to provide training data on where tall and short crops were growing each month, while S2 features were used to map tall crops for all cropland pixels each month. Sugarcane was then identified by leveraging the fact that, among all non-tree species grown in cropland areas, sugarcane is typically tall for the largest fraction of time. Comparisons with field data, pre-existing maps, and official government statistics all indicated high precision and high recall of our maps. Agreement with field data at the pixel level exceeded 80 % in most countries, and subnational sugarcane areas from our maps were consistent with government statistics. Exceptions appeared mainly due to problems in underlying cropland masks or due to under-reporting of sugarcane area by governments. The final maps should be useful in studying the various impacts of sugarcane cultivation and producing maps of related outcomes such as sugarcane yields. The dataset is available on Zenodo at https://doi.org/10.5281/zenodo.10871164 (Di Tommaso et al., 2024a).

A predictive model of treatment effectiveness of refractory peritoneal dialysis-related peritonitis in patients with peritoneal dialysis: a single-center observational study in South China

Abstract Background To prevent loss of peritoneal function caused by persistent abdominal inflammation, the guidelines recommend early extubation in patients with refractory peritoneal dialysis-associated peritonitis (rPDAP). In attempt to pinpoint high-risk patient cohorts that did not respond to treatment for refractory peritonitis, we created a model to predict the effectiveness of peritonitis treatment. Methods This observational cohort study included peritoneal dialysis (PD) patients from January 1, 2011 to December 31, 2020. Multivariate logistic regression analysis was used to explore the factors affecting the occurrence and prognosis of rPDAP, and to construct a predictive model for the success of rPDAP treatment. Receiver operator characteristic (ROC) curve, calibration and decision curve were drawn to evaluate the predictive performance of the model. Results A total of 1, 397 cases of peritoneal dialysis-associated peritonitis (PDAP) occurred in our center during the study period, of which 558 cases were diagnosed as rPDAP. The incidence of refractory peritonitis was 0.047 cases/patient-year. In the study, 440 cases with rPDAP were included. Among them 304 cases (69.1%) had been successfully cured, while 136 cases (30.9%) were treatment failure, of which 19 cases (4.3%) died, 85 (19.3%) cases transferred to hemodialysis and 32 cases (7.2%) were relapse/recurrent peritonitis. Dialysate culture results showed 132 (30.0%) cases were infected with gram positive bacteria, 161 (36.6%) gram negative bacteria. Multivariate Logistic regression analysis showed that episodes of peritonitis previously ≤ 3 times were correlated with the better prognosis of rPDAP, but white blood cell (WBC) counts in peritoneal dialysate on the 3rd day of peritonitis or WBC counts on the 5th day ≥ 300 × 106/L, the pathogenic microorganism with gram-negative bacteria, as well as longer duration of PD were associated with poor outcomes. The C-statistical value of the training data set was 0.870 (95%CI: 0.821–0.918). The calibration curve and clinical decision-making curve also proved that this nomogram could accurately predict the success of treatment in patients with refractory peritonitis. Conclusion The nomogram model created through internal verification indicated a strong clinical application value and a high prognostic prediction accuracy for rPDAP.

Deep learning-based acoustic emission source localization in heterogeneous rock media without prior wave velocity information

Abstract Acoustic emission (AE) source localization is crucial for monitoring but often relies on prior information, such as wave velocity and arrival time. This study introduces a novel method for locating AE sources in rocks without such information, addressing challenges posed by heterogeneous sensor arrays. Experiments involving pencil led break (PLB) tests on sandstone cubes collected AE waveforms and their coordinates. A ResNet-50 based deep learning model was developed to correlate the time-frequency spectra of AE with PLB locations, expressed as spatial Gaussian distributions. The model, achieved a 79% prediction accuracy for AE localization in complex environments. While there is room for improvement in training data quantity and diversity, the results validate the model’s effectiveness, particularly in coal mines and tunnel engineering.

Seizure Sources Can Be Imaged from Scalp EEG by Means of Biophysically Constrained Deep Neural Networks.

Seizure localization is important for managing drug-resistant focal epilepsy. Here, the capability of a novel deep learning-based source imaging framework (DeepSIF) for imaging seizure activities from electroencephalogram (EEG) recordings in drug-resistant focal epilepsy patients is investigated. The neural mass model of ictal oscillations is adopted to generate synthetic training data with spatio-temporal-spectra features similar to ictal dynamics. The trained DeepSIF model is rigorously validated using computer simulations and in a cohort of 33 drug-resistant focal epilepsy patients with high-density (76-channel) EEG seizure recordings, by comparing DeepSIF estimates with surgical resection outcome and seizure onset zone (SOZ) . These findings show that the trained DeepSIF model outperforms other methods in estimating the spatial and temporal information of origins of ictal activities. It achieves a high spatial specificity of 96% and a low spatial dispersion of 3.80±5.74mm when compared to the resection region. The source imaging results also demonstrate good coverage of SOZ, with an average distance of 10.89±10.14mm (from the SOZ to the reconstruction). These promising results suggest that DeepSIF has significant potential for advancing noninvasive imaging of the origins of ictal activities in patients with focal epilepsy, aiding management of intractable epilepsy.

Aviation-BERT-NER: Named Entity Recognition for Aviation Safety Reports

This work introduces Aviation-BERT-NER, a Named Entity Recognition (NER) system tailored for aviation safety reports, building on the Aviation-BERT base model developed at the Georgia Institute of Technology’s Aerospace Systems Design Laboratory. This system integrates aviation domain-specific data, including aircraft types, manufacturers, quantities, and aviation terminology, to identify named entities critical for aviation safety analysis. A key innovation of Aviation-BERT-NER is its template-based approach to fine-tuning, which utilizes structured datasets to generate synthetic training data that mirror the complexity of real-world aviation safety reports. This method significantly improves the model’s generalizability and adaptability, enabling rapid updates and customization to meet evolving domain-specific requirements. The development process involved careful data preparation, including the synthesis of entity types and the generation of labeled datasets through template filling. Testing on real-world narratives from the National Transportation Safety Board (NTSB) database highlighted Aviation-BERT-NER’s robustness, with a precision of 95.34%, recall of 94.62%, and F1 score of 94.78% when evaluated over 50 manually annotated (BIO tagged) paragraphs. This work addresses a critical gap in English language NER models for aviation safety, promising substantial improvements in the analysis and understanding of aviation safety reports.

Designing an Effective Job Recommender System based on Embedded Machine Learning Models

The need to automate employment offers to qualify job searchers has gain attentions. For an automated recommendation systems to be used more frequently, a better user-friendly filtering techniques are required. This paper designs an automated process, referred to as “the job recommender”, which focuses on user-centric design and personalization for recommending and matching applicants with appropriate jobs. We use the bottom-up approach that uses dataset based on filtering algorithms to predict and make recommendations for job seekers. The algorithm helps the recruiters to produce the list of the résumé that best meets the job descriptions. In this context, the random forest (RF) and support vector machines (SVM) are adopted to train the data. They are supplied personalized information (qualifications, result of aptitude test, age, and work experience) reported on the résumés of individual candidates from the pool of submissions, and the system train data to learn the evolution of job selection by candidates based on these machine learning tools. The algorithm used would help the recruiters to produce the list of the résumé that best meets the job descriptions. The algorithms are designed to recommend personalized items tailored to each user's interests. Under the minimum hardware and software requirements, the job recommender system was implemented in streamlit - a python template, for designing the frontend.

HCCL: Hierarchical Counterfactual Contrastive Learning for Robust Visual Question Answering

Despite most state-of-the-art models having achieved amazing performance in Visual Question Answering (VQA) , they usually utilize biases to answer the question. Recently, some studies synthesize counterfactual training samples to help the model to mitigate the biases. However, these synthetic samples need extra annotations and often contain noises. Moreover, these methods simply add synthetic samples to the training data to train the model with the cross-entropy loss, which cannot make the best use of synthetic samples to mitigate the biases. In this article, to mitigate the biases in VQA more effectively, we propose a Hierarchical Counterfactual Contrastive Learning (HCCL) method. Firstly, to avoid introducing noises and extra annotations, our method automatically masks the unimportant features in original pairs to obtain positive samples and create mismatched question-image pairs as negative samples. Then our method uses feature-level and answer-level contrastive learning to make the original sample close to positive samples in the feature space, while away from negative samples in both feature and answer spaces. In this way, the VQA model can learn the robust multimodal features and focus on both visual and language information to produce the answer. Our HCCL method can be adopted in different baselines, and the experimental results on VQA v2, VQA-CP, and GQA-OOD datasets show that our method is effective in mitigating the biases in VQA, which improves the robustness of the VQA model.

Applying Transfer Learning to Improve the Performance of Deep Learning–based Groundwater Level Prediction Model with Insufficient Training Data

Applying Transfer Learning to Improve the Performance of Deep Learning–based Groundwater Level Prediction Model with Insufficient Training Data

Review of Convolutional Neural Network

Over the past decade, the domain of deep learning has emerged as a swiftly expanding area of interest among researchers globally. It offers substantial benefits over traditional shallow networks, particularly in the realms of feature extraction and model fitting. Deep learning excels at uncovering intricate, distributed features from raw data, which exhibit robust generalization capabilities. It has triumphantly addressed challenges that were once deemed intractable within the field of artificial intelligence. With the exponential growth in the volume of training data and a marked increase in computational power, deep learning has achieved remarkable strides and found applications across a spectrum of domains, including but not limited to object detection, computer vision, natural language processing, speech recognition, and semantic parsing, thereby accelerating the evolution of artificial intelligence. Deep learning encompasses a series of non-linear transformations organized hierarchically, with deep neural networks representing the predominant form of contemporary deep learning methodologies. Inspired by the neural connections found in the animal visual cortex, these networks are characterized by local connectivity, shared weights, and pooling operations. Such attributes not only simplify the network model and curtail the number of trainable parameters but also engender invariance to shifts, distortions, and scaling, thereby endowing the model with enhanced robustness and fault tolerance. This makes the training and optimization of the network structure more manageable. This paper commences with a retrospective on the evolution of convolutional neural networks (CNNs). Subsequently, it delineates the architectures of neuron models and multilayer perceptron. It proceeds to dissect the CNN architecture, which typically encompasses a sequence of convolutional and pooling layers, culminating in fully connected layers, each serving distinct functions. The paper also elaborates on various enhanced algorithms for CNNs, such as the “Network in Network” and “Spatial Transformer Networks,” while also introducing supervised and unsupervised learning methodologies pertinent to CNNs and highlighting several prevalent open-source tools. Further, the paper scrutinizes the application of CNNs in various fields, including image classification, facial recognition, audio retrieval, electrocardiogram analysis, and object detection. It posits the amalgamation of CNNs with recurrent neural networks as a potential alternative for training datasets. The research culminates in the design of CNN structures with varying parameters and depths, supported by experimental analysis that elucidates the interplay among these parameters and the ramifications of their configuration. Ultimately, the paper synthesizes the merits and lingering challenges associated with CNNs and their practical applications.

Geometric deep learning framework for de novo genome assembly.

The critical stage of every de novo genome assembler is identifying paths in assembly graphs that correspond to the reconstructed genomic sequences. The existing algorithmic methods struggle with this, primarily due to repetitive regions causing complex graph tangles, leading to fragmented assemblies. Here, we introduce GNNome, a framework for path identification based on geometric deep learning that enables training models on assembly graphs without relying on existing assembly strategies. By leveraging only the symmetries inherent to the problem, GNNome reconstructs assemblies from PacBio HiFi reads with contiguity and quality comparable to those of the state-of-the-art tools across several species. With every new genome assembled telomere-to-telomere, the amount of reliable training data at our disposal increases. Combining the straightforward generation of abundant simulated data for diverse genomic structures with the AI approach makes the proposed framework a plausible cornerstone for future work on reconstructing complex genomes with different ploidy and aneuploidy degrees. To facilitate such developments, we make the framework and the best-performing model publicly available, provided as a tool that can directly be used to assemble new haploid genomes.

Federated learning-based wavelength demodulation system for multi-point distributed multi-peak FBG sensors

Machine learning-based demodulation of multi-peak fiber Bragg grating (FBG) sensors has been extensively studied, demonstrating superior performance compared to conventional algorithms because it can neglect potential physical constraints. As the number of real-world monitoring points increases, the volume of fiber-optic sensing data grows exponentially. This necessitates aggregating data from various regions (e.g., via Wi-Fi), unlike traditional single-point monitoring, which challenges server storage capacity and communication efficiency. To address these issues, this paper proposes a federated learning (FL)-based framework for efficient wavelength demodulation of multi-peak FBGs in multipoint monitoring. Specifically, an arrayed waveguide grating (AWG) with multiplexing capability is employed at different monitoring points to convert spectral features into multi-channel transmission intensities, serving as training data for local models. Subsequently, the local model parameters, trained independently at each point, are uploaded to a central server to derive the optimal global model for demodulation across different monitoring points. The proposed demodulation framework is validated through stress demodulation experiments on multi-peak FBG sensors. Experimental results indicate strong multi-peak extraction performance with a demodulation error of ±0.64 pm. Additionally, the method demonstrates excellent applicability for generating high-precision global demodulation models through multi-node cooperative work under various scenarios.

Emerging Technologies and Applications of AI Chips: Integrating Deep Learning Algorithms with Advanced Hardware Architectures

Due to the ongoing promotion of the latest technological revolution and industrial upgrading, as well as the widespread use of artificial intelligence and deep learning, the intelligent Internet of Things terminal has reached a more advanced stage of development. The combination of training data, model algorithms, and processing power is the basis for advancing artificial intelligence. The semiconductor business plays a crucial role in shaping the future of AI algorithms and the computing power industry. This study explores the progression of artificial intelligence, and essential technologies, and examines the categorization and structure of AI chips. The process of deep learning was then analyzed, demonstrating the application of AI in real life. Next, this study showcases distinct AI chip-related models, specifically designed for the automated implementation of Convolutional Neural Networks (CNNs) on Field-Programmable Gate Arrays (FPGAs). The study then proceeds to examine the unique procedure and the significance of the resulting data. The future of AI was examined in the domains of space exploration, healthcare, and autonomous driving, along with future forecasts.

Together we can do it! A roadmap to effectively tackle propaganda-related tasks

PurposeIn this paper, we address the need to study automatic propaganda detection to establish a course of action when faced with such a complex task. Although many isolated tasks have been proposed, a roadmap on how to best approach a new task from the perspective of text formality or the leverage of existing resources has not been explored yet.Design/methodology/approachWe present a comprehensive study using several datasets on textual propaganda and different techniques to tackle it. We explore diverse collections with varied characteristics and analyze methodologies, from classic machine learning algorithms, to multi-task learning to utilize the available data in such models.FindingsOur results show that transformer-based approaches are the best option with high-quality collections, and emotionally enriched inputs improve the results for Twitter content. Additionally, MTL achieves the best results in two of the five scenarios we analyzed. Notably, in one of the scenarios, the model achieves an F1 score of 0.78, significantly surpassing the transformer baseline model’s F1 score of 0.68.Research limitations/implicationsAfter finding a positive impact when leveraging propaganda’s emotional content, we propose further research into exploiting other complex dimensions, such as moral issues or logical reasoning.Originality/valueBased on our findings, we provide a roadmap for tackling propaganda-related tasks, depending on the types of training data available and the task to solve. This includes the application of MTL, which has yet to be fully exploited in propaganda detection.

Online adaptive critic designs with tensor product B-splines and incremental model techniques

As an effective optimal control scheme in the field of reinforcement learning, adaptive dynamic programming (ADP) has attracted extensive attention in recent decades. Neural networks (NNs) are commonly employed in ADP algorithms to realize nonlinear function approximation. However, the learning process with NN approximation typically requires a substantial amount of training data and places a heavy computational burden. To improve learning efficiency and alleviate computational load, this paper develops a novel model-free ADP approach based on multivariate splines and incremental control techniques, aimed at achieving online learning of nonlinear control. By utilizing input and output data, an incremental linear approximation model is identified online without any prior knowledge of system dynamics. To improve nonlinear approximation capabilities and lessen computational demands, tensor product B-splines, instead of NNs, are integrated into the critic module to approximate the value function, and the recursive least squares temporal difference (RLS-TD) algorithm is employed to update the weights of spline bases. Convergence analysis of the proposed control scheme is conducted based on the dual-timescale stochastic approximation theory. To illustrate the effectiveness and feasibility of the proposed control scheme, numerical simulations are performed in solving the online nonlinear control problem within the context of the inverted pendulum system.

Redefining the accumulated temperature index for accurate prediction of rice flowering time in diverse environments.

Accurate prediction of flowering time across diverse environments is crucial for effective crop management and breeding. While the accumulated temperature index (ATI) is widely used as an indicator for estimating flowering time, its traditional definition lacks systematic evaluation and genetic basis understanding. Here, using data from 422 rice hybrids across 47 locations, we identified the optimal ATI calculation window as 1 day after sowing to 26 days before flowering. Based on this redefined ATI, we developed a single-parameter model that outperforms the state-of-the-art reaction norm index model in both accuracy and stability, especially with limited training data. We identified 10 loci significantly associated with ATI variation, including two near known flowering time genes and four linked to ecotype differentiation. To enhance practical utility, we developed an efficient flowering time prediction kit using 28 functionally relevant markers, complemented by a user-friendly online tool (http://xielab.hzau.edu.cn/ATI). Our approach can be easily applied to other crops, as ATI is commonly used across various agricultural systems.

SPINEPS—automatic whole spine segmentation of T2-weighted MR images using a two-phase approach to multi-class semantic and instance segmentation

Abstract Objectives Introducing SPINEPS, a deep learning method for semantic and instance segmentation of 14 spinal structures (ten vertebra substructures, intervertebral discs, spinal cord, spinal canal, and sacrum) in whole-body sagittal T2-weighted turbo spin echo images. Material and methods This local ethics committee-approved study utilized a public dataset (train/test 179/39 subjects, 137 female), a German National Cohort (NAKO) subset (train/test 1412/65 subjects, mean age 53, 694 female), and an in-house dataset (test 10 subjects, mean age 70, 5 female). SPINEPS is a semantic segmentation model, followed by a sliding window approach utilizing a second model to create instance masks from the semantic ones. Segmentation evaluation metrics included the Dice score and average symmetrical surface distance (ASSD). Statistical significance was assessed using the Wilcoxon signed-rank test. Results On the public dataset, SPINEPS outperformed a nnUNet baseline on every structure and metric (e.g., an average over vertebra instances: dice 0.933 vs 0.911, p < 0.001, ASSD 0.21 vs 0.435, p < 0.001). SPINEPS trained on automated annotations of the NAKO achieves an average global Dice score of 0.918 on the combined NAKO and in-house test split. Adding the training data from the public dataset outperforms this (average instance-wise Dice score over the vertebra substructures 0.803 vs 0.778, average global Dice score 0.931 vs 0.918). Conclusion SPINEPS offers segmentation of 14 spinal structures in T2w sagittal images. It provides a semantic mask and an instance mask separating the vertebrae and intervertebral discs. This is the first publicly available algorithm to enable this segmentation. Key Points QuestionNo publicly available automatic approach can yield semantic and instance segmentation masks for the whole spine (including posterior elements) in T2-weighted sagittal TSE images. FindingsSegmenting semantically first and then instance-wise outperforms a baseline trained directly on instance segmentation. The developed model produces high-resolution MRI segmentations for the whole spine. Clinical relevanceThis study introduces an automatic approach to whole spine segmentation, including posterior elements, in arbitrary fields of view T2w sagittal MR images, enabling easy biomarker extraction, automatic localization of pathologies and degenerative diseases, and quantifying analyses as downstream research.

Relation between wearable heart rate monitor derived training load and cardiac adaption in endurance athletes

Abstract Background No studies have effectively measured training load (TL) from wearable heart rate (HR) monitors and determined its relation to exercise induced left and right ventricular (LV, RV) remodeling. A better insight into drivers of exercise induced cardiac remodeling is of clinical relevance to guide training advice in athletes at risk of adverse remodeling. Purpose We determined the TL, based on HR monitoring by commercial wearables[1]. The number and duration of training sessions was quantified during a period of 3 months. Training-intensity was calculated as Edwards TRIMP (eTRIMP) based on the product of training duration in one of 5 HR zones (1=light, 5=very vigorous) and the coefficient related to the respective HR zone. Methods Training data was collected in 2 prospective multicenter studies. Master@Heart comprises endurance trained middle-aged men aged 45 to 70years[2]. Pro@Heart consists of elite athletes aged 16 to 23years[3]. For this analysis, all male athletes with complete training data recorded using a chest-worn HR monitor for 3 months were included. Using an in-house developed pipeline, maximal HR and TL were calculated from raw training data. Cardiac magnetic resonance imaging was performed to measure indexed LV/RV end-diastolic volume (LVEDVi, RVEDVi), end-systolic volume (LVESVi, RVESVi), ejection fraction (LVEF, RVEF) and LV mass (LVMi). Results Pro@Heart athletes (n=69; age 21.9 ± 4.2 years) trained more than Master@Heart athletes (n=82; age 55.2 ± 6.5 years) and their total 3-month TL (eTRIMP) was also higher (9718 ± 3190 vs. 5499 ± 3775min; P<0.05 and 22339 ± 7350 vs. 13438 ± 8880 AU; P< 0.05; Table 1). Considering all athletes, there was a strong correlation between both training duration and eTRIMP and LVEDVi, LVESVi, LVMi, RVEDVi and RVESVi. Training duration in zone 1&2 correlated well with LVEDVi, LVESVi, LVSV, LVMi, RVEDVi and RVESVi (Table 2). We found small significant negative correlations between duration and eTRIMP compared to LVEF and RVEF (Table 2). To filter out individual training factors influencing cardiac remodeling we used a partial least squares method which included time and percentage of time spent in training zones and age. Accordingly, training duration in zone 1&2 and age were the main predictors of cardiac volumes and LVMi. For LVEF, age and time in zone 3 was most predictive, whereas for RVEF zone 3 and to a lesser degree zone 1&2 were the main predictors, irrespective of age. Interestingly, time spent in high intensity zones 4 and 5 was not predictive. Conclusion We prospectively quantified TL in comparison to CMR-derived measures of cardiac remodeling and demonstrated that training duration in low intensity training zones is most predictive for cardiac volumes and mass, whereas LVEF and RVEF are more closely associated to time in zone 3. Very vigorous intensity exercise was not predictive, likely due to limited time training in zone 4 and 5.

A generalized framework of neural networks for Hamiltonian systems

When solving Hamiltonian systems using numerical integrators, preserving the symplectic structure may be crucial for many problems. At the same time, solving chaotic or stiff problems requires integrators to approximate the trajectories with extreme precision. So, integrating Hamilton's equations to a level of scientific reliability such that the answer can be used for scientific interpretation, may be computationally expensive. However, a neural network can be a viable alternative to numerical integrators, offering high-fidelity solutions orders of magnitudes faster.To understand whether it is also important to preserve the symplecticity when neural networks are used, we analyze three well-known neural network architectures that are including the symplectic structure inside the neural network's topology. Between these neural network architectures many similarities can be found. This allows us to formulate a new, generalized framework for these architectures. In the generalized framework Symplectic Recurrent Neural Networks, SympNets and HénonNets are included as special cases. Additionally, this new framework enables us to find novel neural network topologies by transitioning between the established ones.We compare new Generalized Hamiltonian Neural Networks (GHNNs) against the already established SympNets, HénonNets and physics-unaware multilayer perceptrons. This comparison is performed with data for a pendulum, a double pendulum and a gravitational 3-body problem. In order to achieve a fair comparison, the hyperparameters of the different neural networks are chosen such that the prediction speeds of all four architectures are the same during inference. A special focus lies on the capability of the neural networks to generalize outside the training data. The GHNNs outperform all other neural network architectures for the problems considered.

Temporal relationship between high-intensity training sessions and episodes of atrial fibrillation among previously inactive patients with paroxysmal or persistent AF

Abstract Background Exercise is considered beneficial for managing comorbidities in atrial fibrillation (AF) patients in the long term, and emerging evidence also indicate that high-intensity interval training (HIIT) may reduce AF burden and symptoms. However, more data on the impact of individual training sessions on recurrence of AF is needed to understand mechanisms and tailor exercise recommendations. Purpose To assess the temporal relationship between training sessions and recurrent AF episodes among previously inactive patients with paroxysmal or persistent AF being introduced to HIIT. Methods The Norwegian Exercise in Atrial Fibrillation Trial is a multi centre randomized controlled trial including inactive patients with paroxysmal or persistent AF, who are randomly assigned to either a structured exercise intervention or control for 12 months. We analysed AF episodes and training data from the initial weeks with introduction to structured exercise in the intervention group (8 supervised sessions of HIIT at 85-95% of maximal heart rate). AF episodes lasting ≥30 seconds were registered using implantable cardiac monitors (ICM). Training sessions were registered using sports watches and manual logs. We report event rates and rate ratios for recurrent AF episodes during HIIT compared to episodes occurring outside HIIT. Due to non-normal data distribution statistical comparisons were performed using the Wilcoxon sign-ranked test. Results 81 patients (24 female, 57 male), mean age of 64.2 years (range 35-78), were included. 57 patients had paroxysmal and 24 had persistent AF. The combined duration of monitoring was 5272 days (median 53, range 30 – 163). The participants performed a total of 418 hours of HIIT (median 5.3 hours, range 0.67 – 6.58) during 615 training sessions (median 8, range 1 – 8). In total 4528 AF episodes (median 4, range 0 – 1693) were registered, of which 18 occurred during HIIT. The mean AF rate per 24 hours was 0.69 outside HIIT, 1.20 during HIIT, 0.92 for the periods 1 – 3 hours and 4-6 hours after HIIT (Figure 1). The rate ratio for AF during HIIT compared to outside HIIT was 1.73 (p=0.002). In the 52 participants with at least one episode of AF during follow-up, the rate ratio for AF during HIIT compared to outside HIIT was 10.4 (bootstrapped 95% CI 0.2 to 30.3). Conclusion Among previously inactive patients with paroxysmal or persistent AF, the rate of recurrent AF episodes was significantly higher during high-intensity exercise than outside exercise. Given the few AF episodes, this mechanism of trigging AF by high-intensity exercise needs further investigation.Mean AF per 24 hours