Synthetic Data Research Articles

SeqInfo-SAWGAN-GP: Adaptive feature extraction from vibration time data under variable operating conditions for imbalanced bearing fault diagnosis

Current generative adversarial network (GAN)-based methods are widely used to address data imbalance in intelligent fault diagnosis of bearings, a problem often caused by the scarcity of faulty data in real-world scenarios. However, these methods still face challenges, including high model training difficulty and poor quality under variable operating conditions. To address this, sequence information incorporated self-attention Wasserstein generative adversarial network with gradient penalty (SeqInfo-SAWGAN-GP) is proposed. Firstly, a self-attention block is integrated to extract significant features from limited time data. Then, SeqInfo block is introduced to enhance the quality of synthetic time data and manage the model to generate data across different operating conditions. Additionally, healthy vibration signals corresponding to specific operating conditions are input into the generator to further expedite the generation of synthetic fault data. Finally, the superiority of the proposed method is validated on two bearing datasets. The results indicate that the proposed method achieves accuracy of 98.5% and 95% using only a few samples, marking improvements of 4% and 45% over the original imbalanced dataset on two bearing datasets under variable operating conditions, respectively, and outperforming state-of-the-art methods.

Improving microstructures segmentation via pretraining with synthetic data

Improving microstructures segmentation via pretraining with synthetic data

Spatial unfolding of an extended La-140 source within a 0.5 km-wide exclusion zone using Compton gamma imaging measurements.

Results from applying an advanced spatial-unfolding technique to outdoor-trial data acquired with a Silicon photomultiplier-based Compton Telescope for Safety and Security (SCoTSS) gamma imager during perimeter survey of a distributed La-140 source lying within a 500 m x 500 m exclusion zone are presented. A synthetic-data version of the experiment was also modelled using Monte Carlo simulations and reconstructed. For both experimental and synthetic data the method faithfully reproduces the shape of the activity distribution, and for synthetic data the total activity is reproduced as well. However, for the experimental data the method underestimates the total activity by a factor of six. This can be accounted for by environmental snow and soil conditions not included in the detector response functions. This application of the Compton gamma imager survey-data inversion method demonstrates its applicability under austere conditions wherein extreme weather and transportation constraints severely impacted the quality of the data collected. The trial shows that the method has widespread applicability in the radiological and nuclear safety and security field, particularly for scenarios in which a threat material or contaminated area lies within a no-entry or no-fly zone.

ChatEMG: Synthetic Data Generation to Control a Robotic Hand Orthosis for Stroke.

Intent inferral on a hand orthosis for stroke patients is challenging due to the difficulty of data collection. Additionally, EMG signals exhibit significant variations across different conditions, sessions, and subjects, making it hard for classifiers to generalize. Traditional approaches require a large labeled dataset from the new condition, session, or subject to train intent classifiers; however, this data collection process is burdensome and time-consuming. In this paper, we propose ChatEMG, an autoregressive generative model that can generate synthetic EMG signals conditioned on prompts (i.e., a given sequence of EMG signals). ChatEMG enables us to collect only a small dataset from the new condition, session, or subject and expand it with synthetic samples conditioned on prompts from this new context. ChatEMG leverages a vast repository of previous data via generative training while still remaining context-specific via prompting. Our experiments show that these synthetic samples are classifier-agnostic and can improve intent inferral accuracy for different types of classifiers. We demonstrate that our complete approach can be integrated into a single patient session, including the use of the classifier for functional orthosis-assisted tasks. To the best of our knowledge, this is the first time an intent classifier trained partially on synthetic data has been deployed for functional control of an orthosis by a stroke survivor.

Class-prior probability estimation using density ratio between unlabeled instances and positively labeled noisy instances

In this paper, the authors propose a new approach for determining the mixture proportions of positive and negative instances in both a small set of positively labeled instances contaminated with an unknown proportion of negative instances and an unlabeled instance set with an unknown mixture of positive and negative instances. The unique advantage of this method lies in its ability to derive these proportions with a high level of accuracy by directly leveraging probability density ratio estimation. Experiments conducted on synthetic and UCI repository data demonstrate that the proposed technique can robustly estimate class-prior probabilities with greater speed and better accuracy compared to existing methods, particularly in the presence of instance number and imbalance challenges.

"Play by play": A dataset of handball and basketball game situations in a standardized space.

This paper presents a synthetic dataset of labeled game situations in recordings of federated handball and basketball matches played in Galicia, Spain. The dataset consists of synthetic data generated from real video frames, including 308,805 labeled handball frames and 56,578 labeled basketball frames extracted from 2105 handball and 383 basketball 5-s video clips. Experts manually labeled the video clips based on the respective sports, while the individual frames were automatically labeled using computer vision and machine learning techniques. The dataset encompasses seven classes of game situations: left attack, left counterattack, left penalty, right attack, right counterattack, right penalty, and timeout. In basketball, the penalty class refers to the free throws attempted by players after they have been fouled by an opposing player. Each frame in the dataset is assigned to one of these classes, considering the game situation and specific context. Importantly, the dataset does not contain actual video frames; instead, it provides a synthetic, normalized representation of each frame in JSON format. This tabular data includes player, referee, and ball positions on a normalized field, player and referee velocities, and key regions on the court. Positions of players, referees, and the ball were automatically inferred in each frame by an object detector, followed by a tracking step to detect object positions across frames and compute the velocity vectors. Finally, the obtained coordinates underwent normalization through a perspective transformation, ensuring that the data remained unaffected by variations in camera configurations across different arenas and camera setups. We refer to this standardized coordinate space as the 'unified space'. The dataset holds significant potential for reuse in various domains related to sports analytics and machine learning research. It can serve as a valuable resource for researchers, coaches, and sports enthusiasts, contributing to improvements in player performance, game strategies, match retransmissions, and sports-related technologies.

A novel nonlinear prestack inversion method based on nutcracker optimization algorithm with high convergence speed

Predicting reservoir parameters with high accuracy is still a crucial work of oil reservoir exploration and development. Due to the limitation of computational efficiency, deterministic methods are primarily used in practical production applications for predicting reservoir parameters. When the nonlinear forward equations are exceptionally complex and the initial model constructed deviates significantly from the true reservoir parameters, deterministic methods may have difficulty obtaining reasonable predictions of reservoir parameters. Compared to deterministic methods, intelligent optimization methods based on nature-inspired metaheuristic algorithms have unique advantages because they do not require derivative information, can achieve global optimization, and have less reliance on initial model. Therefore, they perform better in solving complex nonlinear optimization problems. In this paper, a new intelligent optimization algorithm called Nutcracker Optimization Algorithm (NOA) with a high convergence speed is introduced. By utilizing this optimization algorithm to solve the nonlinear inversion problem constructed by the highly nonlinear exact Zoeppritz equations, we analyze the potential of nonlinear reservoir parameters prediction methods based on intelligent optimization algorithms in practical production applications. The synthetic data test shows that, compared to the classical quantum particle swarm optimization (QPSO) algorithm and the highly-cited whale optimization algorithm (WOA), the prestack nonlinear inversion method based on NOA proposed in this paper ensures high convergence accuracy and exhibits high computational efficiency. It significantly reduces computation time and holds great potential for practical production applications. The field data test shows that the proposed method can rapidly and accurately estimates reservoir parameters, validating the feasibility and effectiveness of the proposed method. This has important theoretical value and practical significance for advancing the application of intelligent optimization algorithms in field reservoir exploration and development.

Kinetics, thresholds, and a comparison of mechanisms underlying systemic infection by Listeria monocytogenes

Studies on the system-scale pathogenesis of Listeria monocytogenes infection have classically focused on its ability to colonize in the intestines following an exposure event. However, despite this, many of the most dangerous complications arising from L. monocytogenes infection are observed days, weeks, or months after exposure, resulting indirectly from bacteria escaping this intestinal colonization hub and invading other organs. Over time, findings of various individual phenomena observed during systemic infection have accumulated, including a shift away from the principal route of intestinal dissemination, delays in bacterial colonization of the central nervous system, differing bacterial flux rates across organs, and multi-stability of bacterial population levels. To further our quantitative understanding of foodborne bacterial infection dynamics, a compartmental model of systemic infection which synthesizes these findings is proposed. Under parameterization to infection in BALB/c mice, the model is used to show a substantial decrease in bacterial populations resulting from dissemination through the mesenteric lymph nodes, as compared to the portal vein, when controlling for the number of bacteria passing through each route. Due to the compartmental nature of this model, we anticipate that this result may be paralleled in other microbes which make use of these pathways to escape the intestinal environment. Additionally, we predict thresholds for intestinal dissemination along each of these routes which must be surpassed to induce systemic infection and describe how these thresholds change over time. Supplementarily, logistic curves are fitted to synthetic data as a means of robustly quantifying the dose–response relationship beyond the intestinal barrier.

A pose estimation approach for discarded stacked smartphones recycling: Based on instance segmentation and point cloud registration.

With the rapid increase in end-of-life smartphones, enhancing the automation and intelligence of their recycling processes has become an urgent challenge. At present, the disassembly of discarded smartphones predominantly relies on manual labor, which is not only inefficient but also associated with environmental pollution and high labor intensity. In the context of end-of-life smartphone recycling, complex situations such as stacking and occlusion are commonly encountered. Accurate pose information can provide critical data for precise robotic grasping, thereby improving the level of automation and efficiency in recycling and disassembly. This research proposes a pose estimation method tailored for stacked discarded smartphones, integrating an improved Mask R-CNN instance segmentation model with Iterative Closest Point (ICP) point cloud registration technology. The method begins by accurately segmenting stacked smartphones using both real and synthetic datasets. Subsequently, pose information is extracted through the proposed estimation approach, providing critical data to guide the robotic arm's grasping actions, thereby improving sorting efficiency and minimizing manual intervention. To enhance its practical applicability, a pose recognition interactive system is developed, enabling visualization and dynamic interaction with pose data. Experimental results demonstrate the effectiveness of the transfer learning algorithm, which leverages a large volume of synthetic data combined with a small batch of real-world data. This research offers valuable theoretical insights and technical solutions for advancing the automation and intelligent disassembly of end-of-life smartphones.

AI-ML Based Intelligent De-Smoking / De-Hazing Algorithm

Abstract: This paper presents an AI-ML-based defogging and smoke removal algorithm that uses a hybrid model combined with a convolutional neural network (CNN) and Transformer. The proposed model uses the U-Net backbone to extract local detailed information, uses the FFA-Net maintenance mechanism to update regions with different burning conditions, and uses the Dehamer transformer encoder to capture and manage the comprehensive information of remote locations in the image. This combination works well in a wide range of climate and smoke conditions, producing high-quality images with accurate colors. The model is evaluated on synthetic and real data, yielding results that outperform today's traditional methods and deep learning models. Its effective capabilities make it suitable for immediate use, making it possible to eliminate and eliminate smoke technology in difficult situations.

The spectrum of synthetic data generation: A comprehensive review

Synthetic data, which is the data produced to mimic the characteristics of actual data without revealing any confidential information, is a much safer option than original data, especially when it comes to extreme instances such as personal data, financial data, or military intelligence. There are substantial dangers connected with the use of real-life data such as assault with the intent to commit identity theft, fraud, and hacking, but because synthetic data (SD) reproduces some of the elements of real data, without infringing on anyone’s privacy, suffers from these risks. The project concentrates on the cutting-edge fields of Language Learning Models (LLM) and Deep Learning (DL) to generate synthetic data that mimics real-world data in its intricacy. Advances in LSTM networks and Generative Adversarial Networks (GAN) produce plausible and useful data in sequence forms for natural language processing and machine learning (ML) augmentation respectively. Applications of this technology include, but are not limited to, the use of augmented datasets to improve medical diagnosis, advanced finance fraud detection systems, and designing fictitious consumers in order to enhance AI-based system recommendations. The project which is implemented with Python programming language and also takes advantage of some open source packages such as SymPy, Pydbgen, Synthetic Data Vault (SDV), and Scikit-learn offers a solution to data scarcity and quality problems in order to improve the performance of the AI models in various sectors.

Design and implementation of machine learning-based anomaly detection in iron structures using synthetic data

This manuscript introduces a novel methodology for detecting anomalies in iron structures using synthetic data and machine learning algorithms. Synthetic datasets representing normal and anomalous conditions were generated through simulated gamma-ray interactions with iron. Decision tree and support vector machine (SVM)-based classifiers were employed to train a model capable of distinguishing between intact and defective materials. This data-driven approach provides a scalable and efficient platform for non-destructive testing across industries such as construction, transportation, and manufacturing. In the future, we plan to integrate IoT devices into this framework to enhance its practical applicability. The manuscript presents the design and proposed methodology for machine learning-based anomaly detection in iron structures using synthetic data.

Generative artificial intelligence enables the generation of bone scintigraphy images and improves generalization of deep learning models in data-constrained environments.

Advancements of deep learning in medical imaging are often constrained by the limited availability of large, annotated datasets, resulting in underperforming models when deployed under real-world conditions. This study investigated a generative artificial intelligence (AI) approach to create synthetic medical images taking the example of bone scintigraphy scans, to increase the data diversity of small-scale datasets for more effective model training and improved generalization. We trained a generative model on 99mTc-bone scintigraphy scans from 9,170 patients in one center to generate high-quality and fully anonymized annotated scans of patients representing two distinct disease patterns: abnormal uptake indicative of (i) bone metastases and (ii) cardiac uptake indicative of cardiac amyloidosis. A blinded reader study was performed to assess the clinical validity and quality of the generated data. We investigated the added value of the generated data by augmenting an independent small single-center dataset with synthetic data and by training a deep learning model to detect abnormal uptake in a downstream classification task. We tested this model on 7,472 scans from 6,448 patients across four external sites in a cross-tracer and cross-scanner setting and associated the resulting model predictions with clinical outcomes. The clinical value and high quality of the synthetic imaging data were confirmed by four readers, who were unable to distinguish synthetic scans from real scans (average accuracy: 0.48% [95% CI 0.46-0.51]), disagreeing in 239 (60%) of 400 cases (Fleiss' kappa: 0.18). Adding synthetic data to the training set improved model performance by a mean (± SD) of 33(± 10)% AUC (p < 0.0001) for detecting abnormal uptake indicative of bone metastases and by 5(± 4)% AUC (p < 0.0001) for detecting uptake indicative of cardiac amyloidosis across both internal and external testing cohorts, compared to models without synthetic training data. Patients with predicted abnormal uptake had adverse clinical outcomes (log-rank: p < 0.0001). Generative AI enables the targeted generation of bone scintigraphy images representing different clinical conditions. Our findings point to the potential of synthetic data to overcome challenges in data sharing and in developing reliable and prognostic deep learning models in data-limited environments.

Leveraging synthetic data to improve regional sea level predictions

The rapid increase in sea levels driven by climate change presents serious risks to coastal communities around the globe. Traditional prediction models frequently concentrate on developed regions with extensive tide gauge networks, leaving a significant gap in data and forecasts for developing countries where the tide gauges are sparse. This study presents a novel deep learning approach that combines TimesGAN with ConvLSTM to enhance regional sea level predictions using the more widely available satellite altimetry data. By generating synthetic training data with TimesGAN, we can significantly improve the predictive accuracy of the ConvLSTM model. Our method is tested across three developed regions—Shanghai, New York, and Lisbon—and three developing regions—Liberia, Gabon, and Somalia. The results reveal that integrating TimesGAN reduces the average mean squared error of the ConvLSTM prediction by approximately 66.1%, 76.6%, 64.5%, 78.2%, 81.7% and 85.1% for Shanghai, New York, Lisbon, Liberia, Gabon, and Somalia, respectively. This underscores the effectiveness of synthetic data in enhancing sea level prediction accuracy, across all regions studied.

Sub-cellular population imaging tools reveal stable apical dendrites in hippocampal area CA3

Apical and basal dendrites of pyramidal neurons receive anatomically and functionally distinct inputs, implying compartment-level functional diversity during behavior. To test this, we imaged in vivo calcium signals from soma, apical dendrites, and basal dendrites in mouse hippocampal CA3 pyramidal neurons during head-fixed navigation. To capture compartment-specific population dynamics, we developed computational tools to automatically segment dendrites and extract accurate fluorescence traces from densely labeled neurons. We validated the method on sparsely labeled preparations and synthetic data, predicting an optimal labeling density for high experimental throughput and analytical accuracy. Our method detected rapid, local dendritic activity. Dendrites showed robust spatial tuning, similar to soma but with higher activity rates. Across days, apical dendrites remained more stable and outperformed in decoding of the animal’s position. Thus, population-level apical and basal dendritic differences may reflect distinct compartment-specific input-output functions and computations in CA3. These tools will facilitate future studies mapping sub-cellular activity and their relation to behavior.

A scoping review of privacy and utility metrics in medical synthetic data

The use of synthetic data is a promising solution to facilitate the sharing and reuse of health-related data beyond its initial collection while addressing privacy concerns. However, there is still no consensus on a standardized approach for systematically evaluating the privacy and utility of synthetic data, impeding its broader adoption. In this work, we present a comprehensive review and systematization of current methods for evaluating synthetic health-related data, focusing on both privacy and utility aspects. Our findings suggest that there are a variety of methods for assessing the utility of synthetic data, but no consensus on which method is optimal in which scenario. Moreover, we found that most studies included in this review do not evaluate the privacy protection provided by synthetic data, and those that do often significantly underestimate the risks.

Integrating representation learning, permutation, and optimization to detect lineage-related gene expression patterns

Recent barcoding technologies allow reconstructing lineage trees while capturing paired single-cell RNA-sequencing (scRNA-seq) data. Such datasets provide opportunities to compare gene expression memory maintenance through lineage branching and pinpoint critical genes in these processes. Here we develop Permutation, Optimization, and Representation learning based single Cell gene Expression and Lineage ANalysis (PORCELAN) to identify lineage-informative genes or subtrees where lineage and expression are tightly coupled. We validate our method using synthetic data and apply it to recent paired lineage and scRNA-seq data of lung cancer in a mouse model and embryogenesis of mouse and C. elegans. Our method pinpoints subtrees giving rise to metastases or new cell states, and genes identified as most informative about lineage overlap with known pathways involved in lung cancer progression. Furthermore, our method highlights differences in how gene expression memory is maintained through divisions in cancer and embryogenesis, thereby providing a tool for studying cell state memory through divisions across biological systems.

A dual-domain network with division residual connection and feature fusion for CBCT scatter correction.

This study aims to propose a dual-domain network that not only reduces scatter artifacts but also retains structure details in CBCT. The proposed network comprises a projection-domain sub-network and an image-domain sub-network. The projection-domain sub-network utilizes a division residual network to amplify the difference between scatter signals and imaging signals, facilitating the learning of scatter signals. The image-domain sub-network contains dual encoders and a single decoder. The dual encoders extract features from two inputs parallelly, and the decoder fuses the extracted features from the two encoders and maps the fused features back to the final high-quality image. Of the two input images to the image-domain sub-network, one is the scatter-contaminated image analytically reconstructed from the scatter-contaminated projections, and the other is the pre-processed image reconstructed from the pre-processed projections produced by the projection-domain sub-network. Experimental results on both synthetic and real data demonstrate that our method can effectively reduce scatter artifacts and restore image details. Quantitative analysis using synthetic data shows the mean absolute error (MAE) was reduced by 74% and peak signal-to-noise ratio (PSNR) increased by 57% compared to the scatter-contaminated ones. Testing on real data found a 38% increase in contrast-to-noise ratio (CNR) with our method compared to the scatter-contaminated image. Additionally, our method consistently outperforms comparative methods such as U-Net, DSE-Net, RDCNN and the collimator-based method. A dual-domain network that leverages projection-domain division residual connection and image-domain feature fusion has been proposed for CBCT scatter correction. It has potential applications for reducing scatter artifacts and preserving image details in CBCT.

High-energy Neutrino Source Cross-correlations with Nearest-neighbor Distributions

The astrophysical origins of the majority of the IceCube neutrinos remain unknown. Effectively characterizing the spatial distribution of the neutrino samples and associating the events with astrophysical source catalogs can be challenging given the large atmospheric neutrino background and underlying non-Gaussian spatial features in the neutrino and source samples. In this paper, we investigate a framework for identifying and statistically evaluating the cross-correlations between IceCube data and an astrophysical source catalog based on the k-nearest-neighbor cumulative distribution functions (kNN-CDFs). We propose a maximum likelihood estimation procedure for inferring the true proportions of astrophysical neutrinos in the point-source data. We conduct a statistical power analysis of an associated likelihood ratio test with estimations of its sensitivity and discovery potential with synthetic neutrino data samples and a WISE–2MASS galaxy sample. We apply the method to IceCube’s public ten-year point-source data and find no statistically significant evidence for spatial cross-correlations with the selected galaxy sample. We discuss possible extensions to the current method and explore the method’s potential to identify the cross-correlation signals in data sets with different sample sizes.

Active oversight and quality control in standard Bayesian optimization for autonomous experiments

The fusion of experimental automation and machine learning has catalyzed a new era in materials research, prominently featuring Gaussian Process (GP) Bayesian Optimization (BO) driven autonomous experiments. Here we introduce a Dual-GP approach that enhances traditional GPBO by adding a secondary surrogate model to dynamically constrain the experimental space based on real-time assessments of the raw experimental data. This Dual-GP approach enhances the optimization efficiency of traditional GPBO by isolating more promising space for BO sampling and more valuable experimental data for primary GP training. We also incorporate a flexible, human-in-the-loop intervention method in the Dual-GP workflow to adjust for unanticipated results. We demonstrate the effectiveness of the Dual-GP model with synthetic model data and implement this approach in autonomous pulsed laser deposition experimental data. This Dual-GP approach has broad applicability in diverse GPBO-driven experimental settings, providing a more adaptable and precise framework for refining autonomous experimentation for more efficient optimization.