Homogeneous Dataset Research Articles

Edge-Bound Change Detection in Multisource Remote Sensing Images

Detecting changes in multisource heterogeneous images is a great challenge for unsupervised change detection methods. Image-translation-based methods, which transform two images to be homogeneous for comparison, have become a mainstream approach. However, most of them primarily rely on information from unchanged regions, resulting in networks that cannot fully capture the connection between two heterogeneous representations. Moreover, the lack of a priori information and sufficient training data makes the training vulnerable to the interference of changed pixels. In this paper, we propose an edge-oriented generative adversarial network (EO-GAN) for change detection that indirectly translates images using edge information, which serves as a core and stable link between heterogeneous representations. The EO-GAN is composed of an edge extraction network and a reconstructive network. During the training process, we ensure that the edges extracted from heterogeneous images are as similar as possible through supplemented data based on superpixel segmentation. Experimental results on both heterogeneous and homogeneous datasets demonstrate the effectiveness of our proposed method.

Enhanced Graph Structure Representation for Unsupervised Heterogeneous Change Detection

Heterogeneous change detection (CD) is widely applied in various fields such as urban planning, environmental monitoring, and disaster management. It enhances the accuracy and comprehensiveness of surface change monitoring by integrating multi-sensor remote sensing data. Scholars have proposed many graph-based methods to address the issue of incomparable heterogeneous images caused by imaging differences. However, these methods often overlook the influence of changes in vertex status on the graph structure, which limits their ability to represent image structural features. To tackle this problem, this paper presents an unsupervised heterogeneous CD method based on enhanced graph structure representation (EGSR). This method enhances the representation capacity of the graph structure for image structural features by measuring the unchanged probabilities of vertices, thereby making it easier to detect changes in heterogeneous images. Firstly, we construct the graph structure using image superpixels and measure the structural graph differences of heterogeneous images in the same image domain. Then, we calculate the unchanged probability of each vertex in the structural graph and reconstruct the graph structure using this probability. To accurately represent the graph structure, we adopt an iterative framework for enhancing the representation of the graph structure. Finally, at the end of the iteration, the final change map (CM) is obtained by binary segmentation of the graph vertices based on their unchanged probabilities. The effectiveness of this method is validated through experiments on four sets of heterogeneous image datasets and two sets of homogeneous image datasets.

The PHANGS-AstroSat Atlas of Nearby Star-forming Galaxies

We present the Physics at High Angular resolution in Nearby GalaxieS (PHANGS)-AstroSat atlas, which contains UV imaging of 31 nearby star-forming galaxies captured by the Ultraviolet Imaging Telescope on the AstroSat satellite. The atlas provides a homogeneous data set of far-UV and near-UV maps of galaxies within a distance of 22 Mpc and a median angular resolution of 1.″4 (corresponding to a physical scale between 25 and 160 pc). After subtracting a uniform UV background and accounting for Milky Way extinction, we compare our estimated flux densities to GALEX observations, finding good agreement. We find candidate extended UV disks around the galaxies NGC 6744 and IC 5332. We present the first statistical measurements of the clumping of the UV emission and compare it to the clumping of molecular gas traced with the Atacama Large Millimeter/submillimeter Array (ALMA). We find that bars and spiral arms exhibit the highest degree of clumping, and the molecular gas is even more clumped than the far-UV (FUV) emission in galaxies. We investigate the variation of the ratio of observed FUV to Hα in different galactic environments and kiloparsec-sized apertures. We report that ∼65% of the variation of the log10(FUV/Hα) can be described through a combination of dust attenuation with star formation history parameters. The PHANGS-AstroSat atlas enhances the multiwavelength coverage of our sample, offering a detailed perspective on star formation. When integrated with PHANGS data sets from ALMA, the Very Large Telescope-MUSE, the Hubble Space Telescope, and JWST, it develops our comprehensive understanding of attenuation curves and dust attenuation in star-forming galaxies.

An External Validation Study for Automated Segmentation of Vestibular Schwannoma.

To validate how an automated model for vestibular schwannoma (VS) segmentation developed on an external homogeneous dataset performs when applied to internal heterogeneous data. The external dataset comprised 242 patients with previously untreated, sporadic unilateral VS undergoing Gamma Knife radiosurgery, with homogeneous magnetic resonance imaging (MRI) scans. The internal dataset comprised 10 patients from our institution, with heterogeneous MRI scans. An automated VS segmentation model was developed on the external dataset. The model was tested on the internal dataset. Dice score, which measures agreement between ground truth and predicted segmentations. When applied to the internal patient scans, the automated model achieved a mean Dice score of 61% across all 10 images. There were three tumors that were not detected. These tumors were 0.01 ml on average (SD = 0.00 ml). The mean Dice score for the seven tumors that were detected was 87% (SD = 14%). There was one outlier with Dice of 55%-on further review of this scan, it was discovered that hyperintense petrous bone had been included in the tumor segmentation. We show that an automated segmentation model developed using a restrictive set of siloed institutional data can be successfully adapted for data from different imaging systems and patient populations. This is an important step toward the validation of automated VS segmentation. However, there are significant shortcomings that likely reflect limitations of the data used to train the model. Further validation is needed to make automated segmentation for VS generalizable.

Aerosol Optical Properties Retrieved by Polarization Raman Lidar: Methodology and Strategy of a Quality-Assurance Tool

Aerosol optical properties retrieved using polarization Raman lidar observations play an increasingly vital role in meteorology and environmental protection. The quality of the data products directly affects the impact of relevant scientific applications. However, the quality of aerosol optical properties retrieved from polarization Raman lidar signals is difficult to assess. Various factors, such as hardware system performance, retrieval algorithm, and meteorological conditions at the observation site, influence data quality. In this study, we propose a method that allows for assessing the reliability of aerosol optical properties derived from polarization Raman lidar observations. We analyze the factors that affect the reliability of retrieved aerosol optical properties. We use scoring methods combined with a weight-assignment scheme to evaluate the quality of the retrieved aerosol optical properties. The scores and weights of each factor are arranged based on our analysis of a simulation study and the characteristics of each factor. We developed an automatic retrieval algorithm that allows for deriving homogeneous aerosol optical data sets. We also assess with this method the quality of retrieved aerosol optical properties obtained with different polarization Raman lidars under different measurement scenarios. Our results show that the proposed quality assurance method can distinguish the reliability of the retrieved aerosol optical properties.

A catalogue of observed geo-effective CME/ICME characteristics

One of the goals of Space Weather studies is to achieve a better understanding of impulsive phenomena, such as Coronal Mass Ejections (CMEs), to improve our ability to forecast their propagation characteristics and mitigate the risks to our technologically driven society. The essential part of achieving this goal is to assess the performance of forecasting models. To this end, the quality and availability of suitable data are of paramount importance. In this work, we merged publicly available data of CMEs from both in-situ and remote observations in order to build a dataset of CME properties. To evaluate the accuracy of the dataset and confirm the relationship between in-situ and remote observations, we have employed the Drag-Based Model (DBM) due to its simplicity and modest consumption of computational resources. In this study, we have also explored the parameter space for the drag parameter and solar wind speed using a Monte Carlo approach to evaluate how efficiently the DBM determines the propagation of CMEs for the events in the dataset. The geoeffective CMEs selected as a result of this work are compliant with the hypothesis of DBM (isolated CME, constant solar wind speed beyond 20 R⊙) and also yield further insight into CME features such as arrival time and arrival speed at L1 point, lift-off time, speed at 20 R⊙ and other similar quantities. Our analysis based on the acceptance rate in the DBM inversion procedure shows that almost 50% of the CME events in the dataset are well described by DBM as they propagate in the heliosphere. The dataset includes statistical metrics for the DBM model parameters. The probability distribution functions (PDFs) for the free parameters of DBM have been derived through a Monte Carlo-like inversion procedure. Probability distribution functions obtained from this work are comparable to PDFs employed in previous works. The analysis showed that there exist two different most probable values (median values) of solar wind speed for DBM input based on slow (wslow ≈ 386 km/s) and fast (wfast ≈ 547 km/s) solar wind type. The most probable value for the drag parameter (γ ≈ 0.687 × 10−7 km−1) in our study is somewhat higher than the values reported in previous studies. Using a data-driven approach, this procedure allows us to present a homogeneous, reliable, and robust dataset for the investigation of CME propagation. Additionally, possible CME events are identified where the DBM prediction is not valid due to model limitations and higher uncertainties in the input parameters. These events require further thorough investigation in the future.

Analysis of the results of preventive testing of bushings using elements of the theory of statistical hypothesis testing

The article presents the results of a study of the dependence of insulation indicators on the duration of operation for serviceable and faulty high-voltage oil-filled bushings, both of hermetic and non-hermetic construction. The results of periodic tests of oil-filled bushings, with a voltage of 110 kV of hermetic design with paper-oil insulation of the capacitor type, as well as non-hermetic construction, were used as initial data. Three statistical criteria were used to form statistically homogeneous data sets: the Wilcoxon rank test, the Z-test, and the Fisher-Snedecor test. The values of the sample mean, sample variance, as well as coefficients of asymmetry and excess of insulation indicators of high-voltage bushings were obtained for the array of initial data and arrays obtained during statistical processing. The values of the sample averages for the arrays obtained as a result of the statistical processing procedure differ significantly. There is a significant decrease in the sample variance values for tgd1 compared to the original arrays. To check the homogeneity of the data in the obtained arrays, the one-factor variance analysis model is used. The practical use of this algorithm made it possible to significantly reduce the heterogeneity of the results of periodic tests of the state of insulation of high-voltage bushings. The analysis of the composition of the arrays with homogeneous values of the indicators made it possible to establish that the values of the indicators of bushings from different regions of Ukraine, of different types and made with different numbers of factory drawings, got into the homogeneous data arrays. It is noted that for high-voltage bushings that do not have insulation defects, the values of the mathematical expectation of the same indicator for different arrays of preventive tests of capacitor-type insulation have a shift relative to each other. From this it follows that the load factor and the operating time of the transformer significantly affect the determination of the maximum allowable value of the insulation index of the high-voltage bushing.

Classification of tumor types using XGBoost machine learning model: a vector space transformation of genomic alterations

BackgroundMachine learning (ML) represents a powerful tool to capture relationships between molecular alterations and cancer types and to extract biological information. Here, we developed a plain ML model aimed at distinguishing cancer types based on genetic lesions, providing an additional tool to improve cancer diagnosis, particularly for tumors of unknown origin.MethodsTCGA data from 9,927 samples spanning 32 different cancer types were downloaded from cBioportal. A vector space model type data transformation technique was designed to build consistently homogeneous new datasets containing, as predictive features, calls for somatic point mutations and copy number variations at chromosome arm-level, thus allowing the use of the XGBoost classifier models. Considering the imbalance in the dataset, due to large difference in the number of cases for each tumor, two preprocessing strategies were considered: i) setting a percentage cut-off threshold to remove less represented cancer types, ii) dividing cancer types into different groups based on biological criteria and training a specific XGBoost model for each of them. The performance of all trained models was mainly assessed by the out-of-sample balanced accuracy (BACC) and the AUC scores.ResultsThe XGBoost classifier achieved the best performance (BACC 77%; AUC 97%) on a dataset containing the 10 most represented tumor types. Moreover, dividing the 18 most represented cancers into three different groups (endocrine-related carcinomas, other carcinomas and other cancers),such analysis models achieved 78%, 71% and 86% BACC, respectively, with AUC scores greater than 96%. In addition, the model capable of linking each group to a specific cancer type reached 81% BACC and 94% AUC. Overall, the diagnostic potential of our model was comparable/higher with respect to others already described in literature and based on similar molecular data and ML approaches.ConclusionsA boosted ML approach able to accurately discriminate different cancer types was developed. The methodology builds datasets simpler and more interpretable than the original data, while keeping enough information to accurately train standard ML models without resorting to sophisticated Deep Learning architectures. In combination with histopathological examinations, this approach could improve cancer diagnosis by using specific DNA alterations, processed by a replicable and easy-to-use automated technology. The study encourages new investigations which could further increase the classifier’s performance, for example by considering more features and dividing tumors into their main molecular subtypes.

Precision Nitrogen Fertilization for Opium Poppy Using Combined Proximal and Remote Sensor Data Fusion

Proper management of within-field variability is crucial for maximizing crop yield, production outcomes and resource use efficiency and reducing environmental impacts. This study evaluated the agroeconomic and environmental feasibilities of site-specific nitrogen fertilization (SNF) in opium poppy (Papaver somniferum L.). On-line visible and near-infrared reflectance spectroscopy was used to estimate soil pH, organic carbon (OC), soil organic matter (SOM), P, K, Mg, Ca, Na, moisture content (MC), Ca:Mg and K:Mg for one field in Spain. Normalized difference vegetation indexes of the previous crop were retrieved from Sentine-2 images. Rasterization of soil and crop data layers created a spatially homogenous dataset followed by delineation of a management zone (MZ) map using a k-means cluster analysis. MZ clusters were ranked relying on the within-cluster soil fertility attributes. A strip experiment was conducted by creating parallel stripes distributed over the MZ map, over which two SNF treatments (i.e., SNF-Kings approach [KA] and SNF-Robin Hood approach [RHA]) were compared against the uniform rate N (URN) control treatment. In SNF-KA, the highest and lowest N dose was applied in the most and least fertile MZ, respectively, whereas the opposite approach was adopted in the SNF-RHA treatment. Yield and cost–benefit analyses provided both SNF treatments to produce more yield (KA = 2.72 and RHA = 2.74 t ha−1) than the URN (2.64 t ha−1) treatment, leading to increasing gross margins by EUR 91 ha−1 (SNF–KA) and EUR 88.5 ha−1 (SNF–RHA). While SNF-KA reduced N input by 66.54 kg N ha−1, SNF–RHA applied more N by 17.90 kg N ha−1 than URN. Additionally, SNF–RHA attempted to equalize yield responses to N across MZ classes, with a small increase in N input. This study, therefore, suggests adopting SNF–RHA for increasing yield and gross margin and accurate distribution of N according to per MZ N response. Future studies, however, should address the limitations of the current study by delineating MZ maps with the incorporation of additional soil information (e.g., mineral N and clay) for optimizing N doses as well as evaluating agroeconomic performance across multiple sites and years using a full-budget analysis.

Semantic segmentation of plant roots from RGB (mini-) rhizotron images—generalisation potential and false positives of established methods and advanced deep-learning models

BackgroundManual analysis of (mini-)rhizotron (MR) images is tedious. Several methods have been proposed for semantic root segmentation based on homogeneous, single-source MR datasets. Recent advances in deep learning (DL) have enabled automated feature extraction, but comparisons of segmentation accuracy, false positives and transferability are virtually lacking. Here we compare six state-of-the-art methods and propose two improved DL models for semantic root segmentation using a large MR dataset with and without augmented data. We determine the performance of the methods on a homogeneous maize dataset, and a mixed dataset of > 8 species (mixtures), 6 soil types and 4 imaging systems. The generalisation potential of the derived DL models is determined on a distinct, unseen dataset.ResultsThe best performance was achieved by the U-Net models; the more complex the encoder the better the accuracy and generalisation of the model. The heterogeneous mixed MR dataset was a particularly challenging for the non-U-Net techniques. Data augmentation enhanced model performance. We demonstrated the improved performance of deep meta-architectures and feature extractors, and a reduction in the number of false positives.ConclusionsAlthough correction factors are still required to match human labelled root lengths, neural network architectures greatly reduce the time required to compute the root length. The more complex architectures illustrate how future improvements in root segmentation within MR images can be achieved, particularly reaching higher segmentation accuracies and model generalisation when analysing real-world datasets with artefacts—limiting the need for model retraining.

ANN-based ground motion model for Turkey using stochastic simulation of earthquakes

SUMMARY Turkey is characterized by a high level of seismic activity attributed to its complex tectonic structure. The country has a dense network to record earthquake ground motions; however, to study previous earthquakes and to account for potential future ones, ground motion simulations are required. Ground motion simulation techniques offer an alternative means of generating region-specific time-series data for locations with limited seismic networks or regions with seismic data gaps, facilitating the study of potential catastrophic earthquakes. In this research, a local ground motion model (GMM) for Turkey is developed using region-specific simulated records, thus constructing a homogeneous data set. The simulations employ the stochastic finite-fault approach and utilize validated input-model parameters in distinct regions, namely Afyon, Erzincan, Duzce, Istanbul and Van. To overcome the limitations of linear regression-based models, artificial neural network is used to establish the form of equations and coefficients. The predictive input parameters encompass fault mechanism (FM), focal depth (FD), moment magnitude (Mw), Joyner and Boore distance (RJB) and average shear wave velocity in the top 30 m (Vs30). The data set comprises 7359 records with Mw ranging between 5.0 and 7.5 and RJB ranging from 0 to 272 km. The results are presented in terms of spectral ordinates within the period range of 0.03–2.0 s, as well as peak ground acceleration and peak ground velocity. The quantification of the GMM uncertainty is achieved through the analysis of residuals, enabling insights into inter- and intra-event uncertainties. The simulation results and the effectiveness of the model are verified by comparing the predicted values of ground motion parameters with the observed values recorded during previous events in the region. The results demonstrate the efficacy of the proposed model in simulating physical phenomena.

Data Mining and Fusion Framework for In-Home Monitoring Applications.

Sensor Data Fusion (SDT) algorithms and models have been widely used in diverse applications. One of the main challenges of SDT includes how to deal with heterogeneous and complex datasets with different formats. The present work utilised both homogenous and heterogeneous datasets to propose a novel SDT framework. It compares data mining-based fusion software packages such as RapidMiner Studio, Anaconda, Weka, and Orange, and proposes a data fusion framework suitable for in-home applications. A total of 574 privacy-friendly (binary) images and 1722 datasets gleaned from thermal and Radar sensing solutions, respectively, were fused using the software packages on instances of homogeneous and heterogeneous data aggregation. Experimental results indicated that the proposed fusion framework achieved an average Classification Accuracy of 84.7% and 95.7% on homogeneous and heterogeneous datasets, respectively, with the help of data mining and machine learning models such as Naïve Bayes, Decision Tree, Neural Network, Random Forest, Stochastic Gradient Descent, Support Vector Machine, and CN2 Induction. Further evaluation of the Sensor Data Fusion framework based on cross-validation of features indicated average values of 94.4% for Classification Accuracy, 95.7% for Precision, and 96.4% for Recall. The novelty of the proposed framework includes cost and timesaving advantages for data labelling and preparation, and feature extraction.

Assessing 36 Years of the Forced Choice Design in Extra Sensory Perception Research: A Meta-Analysis, 1987 to 2022

This meta-analysis is an update of Storm, Tressoldi, and Di Risio (2012); a meta-analysis on forced-choice ESP studies (1987 to 2010), which use targets such as card symbols, pictures, and letters. We formed two datasets: New Studies #1 (studies that included actual hit rates) and New Studies #2 (Reaction Time [‘RT’] Studies; which are studies that measured only reaction time, not hits, as indicators of psi responses). New Studies #1: For the period 2011 to 2022, a homogeneous dataset of 38 studies yielded a mean effect size (ES) of 0.02 (Stouffer Z = 5.55, p = 1.43 × 10–8). New Studies #2 (‘RT’ Studies): For the same period, a homogeneous dataset of 23 studies yielded a weaker mean ES of 0.01 (Stouffer Z = 5.50, p = 1.90 × 10–8). The two databases were combined. In this dataset, telepathy, clairvoyance, and precognition studies were not significantly different from each other. Nor were target types. We updated the forced-choice database by combining our revised original database with the new studies to form a homogeneous database (N = 141): mean effect size (ES) of 0.02 (Stouffer Z = 8.52, p < 10-16). Effects did not vary between investigators or laboratories, and we found a near-significant incline in ES values over a 36-year period (i.e., no evidence of a decline). These results confirm that the forced-choice design adequately tests extra-sensory perception (ESP). We compare the overall results with those obtained in other domains, focusing on ‘selected’ participants (meditators, psychics, psi-test experienced) and ‘unselected’ (i.e., untrained, naïve) participants.

Call for algorithmic fairness to mitigate amplification of racial biases in artificial intelligence models used in orthodontics and craniofacial health.

Machine Learning (ML), a subfield of Artificial Intelligence (AI), is being increasingly used in Orthodontics and craniofacial health for predicting clinical outcomes. Current ML/AI models are prone to accentuate racial disparities. The objective of this narrative review is to provide an overview of how AI/ML models perpetuate racial biases and how we can mitigate this situation. A narrative review of articles published in the medical literature on racial biases and the use of AI/ML models was undertaken. Current AI/ML models are built on homogenous clinical datasets that have a gross underrepresentation of historically disadvantages demographic groups, especially the ethno-racial minorities. The consequence of such AI/ML models is that they perform poorly when deployed on ethno-racial minorities thus further amplifying racial biases. Healthcare providers, policymakers, AI developers and all stakeholders should pay close attention to various steps in the pipeline of building AI/ML models and every effort must be made to establish algorithmic fairness to redress inequities.

APPLICATION OF 68GA-FAPI PET/CT IN CLINICAL PRACTICE – PERSPECTIVES FOR MALIGNANT TUMOR IMAGING: A LITERATURE REVIEW

Relevance: The incidence of malignant neoplasms of various localizations is growing worldwide and in Kazakhstan. The mortality rate from oncological diseases is also alarmingly high. To facilitate early diagnosis and optimal therapy, scientists are exploring molecular diagnostics, including PET/CT, using various markers, like 18F-fluorodeoxyglucose, widely used in oncology but lacking specificity for certain types of tumors. The finding of Fibroblast Activation Protein (FAP) has sparked interest in FAP-targeted radiolabeled inhibitors (FAPI), which could serve as a universal marker for diagnosing different types of cancer. Various FAP markers for PET/CT are being studied, with special attention given to 68Ga-FAPI.
 The study aimed to analyze the potential value of FAPI PET/CT for detecting malignant tumors.
 Methods: A literature review was conducted using the MEDLINE, Embase, Scopus, PubMed, and Cochrane Central Register of Controlled Trials databases for the past decade using the following keywords: “malignant lesions,” “PET/CT,” and “FAPI.” This review analyzes 48 literature sources with A1-level evidence dedicated to the 68Ga-FAPI PET/CT diagnostic accuracy in detecting and staging malignant tumors and assessing treatment efficacy.
 Results: According to the analyzed sources, the 68Ga-FAPI PET/CT sensitivity and specificity in diagnosing cancer is 95% to 100% and 62% to 100%, respectively. However, clear indications for use in clinical practice require further study of 68Ga-FAPI PET/CT diagnostic capabilities on larger cohorts and more homogeneous datasets.
 Conclusion: The available literature data on FAPI PET/CT diagnostic capacity shows this marker’s potential in diagnosing oncological disorders. Information provided by 68Ga-FAPI PET/CT supplements the existing nuclear medicine methods and generally impacts the treatment strategy for each unique case.

VNAGT: Variational Non-Autoregressive Graph Transformer Network for Multi-Agent Trajectory Prediction

Accurately predicting the trajectory of road agents in complex traffic scenarios is challenging because the movement patterns of agents are complex and stochastic, not only depending on their own past trajectories but also being closely related to the social interaction with other surrounding agents. Besides accuracy, efficient prediction with low inference latency is also a highly desirable feature for the practical application of trajectory prediction. To address these issues, we propose the VNAGT model, a variational non-autoregressive graph transformer that adopts the framework of conditional variational autoencoder and incorporates the non-autoregressive approach such that diverse trajectories and low prediction latency can be achieved simultaneously. In order to capture the social and temporal interaction, we put forward a unified graph attention-based module that is applicable for homogeneous and heterogeneous multi-agents such that the class information can be seamlessly integrated when it is available. Non-autoregressive decoding is combined with variational learning to produce multiple plausible predictions with low latency. We train and validate the model on two real-world homogeneous and heterogeneous trajectory datasets. The experimental results demonstrate the superior performance of our model in comparison with the state-of-the-art methods.

Convolutional auto-encoded extreme learning machine for incremental learning of heterogeneous images

<span lang="EN-US">In real-world scenarios, a system's continual updating of learning knowledge becomes more critical as the data grows faster, producing vast volumes of data. Moreover, the learning process becomes complex when the data features become varied due to the addition or deletion of classes. In such cases, the generated model should learn effectively. Incremental learning refers to the learning of data which constantly arrives over time. This learning requires continuous model adaptation but with limited memory resources without sacrificing model accuracy. In this paper, we proposed a straightforward knowledge transfer algorithm (convolutional auto-encoded extreme learning machine (CAE-ELM)) implemented through the incremental learning methodology for the task of supervised classification using an extreme learning machine (ELM). Incremental learning is achieved by creating an individual train model for each set of homogeneous data and incorporating the knowledge transfer among the models without sacrificing accuracy with minimal memory resources. In CAE-ELM, convolutional neural network (CNN) extracts the features, stacked autoencoder (SAE) reduces the size, and ELM learns and classifies the images. Our proposed algorithm is implemented and experimented on various standard datasets: MNIST, ORL, JAFFE, FERET and Caltech. The results show a positive sign of the correctness of the proposed algorithm.</span>

Mining multi-center heterogeneous medical data with distributed synthetic learning

Overcoming barriers on the use of multi-center data for medical analytics is challenging due to privacy protection and data heterogeneity in the healthcare system. In this study, we propose the Distributed Synthetic Learning (DSL) architecture to learn across multiple medical centers and ensure the protection of sensitive personal information. DSL enables the building of a homogeneous dataset with entirely synthetic medical images via a form of GAN-based synthetic learning. The proposed DSL architecture has the following key functionalities: multi-modality learning, missing modality completion learning, and continual learning. We systematically evaluate the performance of DSL on different medical applications using cardiac computed tomography angiography (CTA), brain tumor MRI, and histopathology nuclei datasets. Extensive experiments demonstrate the superior performance of DSL as a high-quality synthetic medical image provider by the use of an ideal synthetic quality metric called Dist-FID. We show that DSL can be adapted to heterogeneous data and remarkably outperforms the real misaligned modalities segmentation model by 55% and the temporal datasets segmentation model by 8%.

Homogenizing SOHO/EIT and SDO/AIA 171 Å Images: A Deep-learning Approach

Extreme-ultraviolet (EUV) images of the Sun are becoming an integral part of space weather prediction tasks. However, having different surveys requires the development of instrument-specific prediction algorithms. As an alternative, it is possible to combine multiple surveys to create a homogeneous data set. In this study, we utilize the temporal overlap of Solar and Heliospheric Observatory Extreme ultraviolet Imaging Telescope and Solar Dynamics Observatory Atmospheric Imaging Assembly 171 Å surveys to train an ensemble of deep-learning models for creating a single homogeneous survey of EUV images for two solar cycles. Prior applications of deep learning have focused on validating the homogeneity of the output while overlooking the systematic estimation of uncertainty. We use an approach called “approximate Bayesian ensembling” to generate an ensemble of models whose uncertainty mimics that of a fully Bayesian neural network at a fraction of the cost. We find that ensemble uncertainty goes down as the training set size increases. Additionally, we show that the model ensemble adds immense value to the prediction by showing higher uncertainty in test data that are not well represented in the training data.

Stability follows efficiency based on the analysis of a large perovskite solar cells ageing dataset

While perovskite solar cells have reached competitive efficiency values during the last decade, stability issues remain a critical challenge to be addressed for pushing this technology towards commercialisation. In this study, we analyse a large homogeneous dataset of Maximum Power Point Tracking (MPPT) operational ageing data that we collected with a custom-built High-throughput Ageing System in the past 3 years. In total, 2,245 MPPT ageing curves are analysed which were obtained under controlled conditions (continuous illumination, controlled temperature and atmosphere) from devices comprising various lead-halide perovskite absorbers, charge selective layers, contact layers, and architectures. In a high-level statistical analysis, we find a correlation between the maximum reached power conversion efficiency (PCE) and the relative PCE loss observed after 150-hours of ageing, with more efficient cells statistically also showing higher stability. Additionally, using the unsupervised machine learning method self-organising map, we cluster this dataset based on the degradation curve shapes. We find a correlation between the frequency of particular shapes of degradation curves and the maximum reached PCE.