Articles tagged with Data Sets
A Mass General Brigham study highlights inconsistencies in generative AI that can affect patient safety if not addressed. The researchers found 'drift' (model performance changes over time) and 'nondeterminism' (inconsistent results between runs) in their tests, emphasizing the need for repeated testing and monitoring.
Researchers at University of Bath and Technical University of Darmstadt found that large language models like ChatGPT cannot learn independently or acquire new skills, making them controllable and predictable. The study concluded that LLMs remain inherently safe, but misuse is still possible.
A new study found a significant sex bias in pain management at emergency departments, with female patients consistently receiving fewer pain medication prescriptions compared to male patients. Female patients also spend more time in the emergency department and have their pain scores less frequently recorded.
Researchers studied mesoscale eddy observations to estimate theoretical predictability limits of eddy trajectories. Long-lived eddies have higher predictability limits, while short-lived ones are less predictable. The study also introduced a complexity index to elucidate OME track complexity.
A recent study at Rice University found that using synthetic data to train generative AI models can lead to negative consequences, including model collapse and reduced quality. As models become increasingly dependent on self-consuming loops, they may produce warped outputs lacking diversity or quality.
A recent study published in Frontiers in Immunology highlights the crucial role of tissue-resident memory T cells in non-small cell lung cancer. The research found that these cells can significantly impact patient outcomes and guide personalized treatment strategies, particularly those involving immunotherapy.
A new study by Newcastle University shows that citizen scientists can accurately identify slug species with proper training and support. The research found that participants improved their identification skills throughout the project, with accuracy rates ranging from 47-70%.
Researchers found the median peak age for Olympic track-and-field athletes to be 27 years old. After 27, the probability of a peak performance drops significantly, highlighting the importance of training regimens in optimizing results.
The A4 study, a large clinical trial of pre-symptomatic Alzheimer’s disease, has yielded key insights into the condition. Researchers have collected extensive data on brain scans, blood samples, genetic information and cognitive tests from over 7,500 participants.
A new study published in Aging explores the potential of three large DNA methylation datasets to identify biological age signals in dogs. The researchers found that biological age methylation clocks are affected by population stratification and require heavy parameterization to achieve effective predictions.
Researchers developed a deep learning-based model to estimate rainfall intensity from surveillance audio, achieving a root mean absolute error of 0.88 mm h-1 and a coefficient of correlation of 0.765. This approach offers a new method for high-resolution hydrological sensing, contributing to environmental resilience and urban sensing.
Research by University of East Anglia and University of Texas found that stricter data privacy laws significantly reduced breaches, but negatively affected firms' market value. Companies compliant with GDPR invested more in data protection and were less likely to experience data breaches.
Scientists have discovered that planted mangroves can store a significant amount of carbon, reaching levels comparable to those in intact stands after just 20 years. The study used logistic models compiled from over 700 planted mangrove stands worldwide and found that the trees' carbon stock reached 71-73% of that found in intact stands.
Researchers developed a machine learning approach to identify potential subtypes in diseases, significantly enhancing disease classification and treatment strategies. The model uncovered 515 previously unannotated disease subtypes.
Researchers re-analyzed historical data from a classic displacement experiment and found that young starlings migrated independently, using their own direction. Adult starlings adjusted their migratory orientation to reach their normal wintering areas, while local conspecifics influenced the route of relocated young starlings.
A team of researchers developed EMOKINE software to measure the objective kinematic features of movements that express emotions. The software provides movement parameters from data sets at the touch of a button, allowing scientists to analyze emotional expression and intentions.
The e-COL+ project aims to capture and reconstruct France's natural history collections in 3D, covering nearly 6% of the world's total natural specimens. The project will provide modern equipment, create a comprehensive dataset of 3D models, and build AI tools to improve model reconstruction.
The new interactive atlas provides a library of case studies for adapting to drought in the US Southwest. The atlas offers geospatial solutions to water scarcity, helping communities implement effective adaptation strategies.
Researchers developed FairDeDup, a cost-effective method to reduce social biases in AI systems by removing redundant data and incorporating controllable diversity dimensions. The approach enables accurate and fair AI training with fewer resources.
Research by Lancaster University found that people in higher social grades, including corporate world and education sectors, are adopting each other's speech patterns to be more inclusive. This 'resonance' has increased over the past 20 years, particularly among those with high social status.
A new method, PURPLE, estimates how often underreported health conditions occur in different demographic groups, providing a more accurate picture of intimate partner violence. The algorithm shows that women from lower-income brackets are more likely to experience violence, consistent with previous literature.
The Kids First DRC has introduced an upgraded data portal to streamline big data search and analysis, improving collaborative pediatric research outcomes. The new portal integrates diverse datasets, including genomic information from the Children's Brain Tumor Network, to foster cross-disciplinary research.
A comprehensive, user-friendly repository has been created to help study Alzheimer's disease. The ssREAD database encompasses 277 integrated datasets from 67 scRNA-seq & snRNA-seq studies, totaling 7,332,202 cells, and includes interactive visualizations for comprehensive analysis interpretations.
The study uses the Dermatological Vision Dataset and compares vision transformers with traditional CNNs, achieving an impressive 97.8% accuracy on validation sets. Integrating ViTs into dermatology represents a promising step toward more accurate diagnostics.
Researchers developed a reliable iris recognition method by applying statistical limits to the spatial domain zero crossing technique, reducing errors to 0.022%. The algorithm uses a neural network to recognize unique features of each person's iris, achieving over 99.78% accuracy.
Portland State University has secured a nearly $1 million grant from the National Science Foundation's Campus Cyberinfrastructure program to establish the Oregon Regional Computing Accelerator (Orca) cluster. The cluster will provide free-of-cost computing resources and cyberinfrastructure to colleges in rural, regional, and minority-s...
A novel approach to training AI systems uses information about spatial position to identify objects and navigate surroundings, inspired by children's visual development. The method improves contrastive learning models' effectiveness by incorporating simulated spatial context information, outperforming base models in various tasks.
A new dataset integrates global Health AI research, providing a structured resource for researchers, policymakers, and practitioners. The dataset includes 96,332 Health AI documents, covering publications, open research datasets, patents, grants, and clinical trials.
A new study from the University of Notre Dame shows that repeated trips to the grocery store can be a reliable indicator of creditworthiness. The research found that buying healthier groceries and consistent shopping habits are linked to on-time credit card payments.
A new transit station in Japan significantly reduced average healthcare expenditures per capita over four years, with savings of approximately $929.99. The study used a causal impact algorithm and time series data to analyze the medical expenditure data gathered from a suburban city on the West Japan Railway line.
A recent study by Kyoto University reveals that expatriates' boundary-spanning activities can lead to role stress, emotional exhaustion, and a sense of being an outsider among local employees. The study highlights the need for careful management to mitigate these negative impacts.
A team at the University of Münster developed an improved method for explaining machine predictions of chemical reactions, using mechanisms such as reproduction, mutation and selection. The algorithm creates customised molecular fingerprints that predict chemical reactions with surprising accuracy, suitable for predicting quantum chemi...
Research suggests that four proposed AI search engines for automating search and retrieval of digital histopathology slides have inadequate performance for routine clinical care. The algorithms showed less than 50% accuracy in some cases, highlighting the need for rigorous external validations and standardization before clinical adoption.
A study by Washington State University found ChatGPT's generative AI system provided inconsistent heart risk assessments for patients with chest pain. The AI failed to match traditional methods used by physicians and returned different results for the same patient data, highlighting its limitations in high-stakes clinical situations.
The new 'scLENS' tool overcomes challenges in single-cell transcriptomics by automatically differentiating signals from noise using Random Matrix Theory and Signal robustness test. This innovation significantly improves analysis accuracy and efficiency, enabling researchers to extract biological signals conveniently and automatically.
A new AI model generates realistic images of single cells, which are used as synthetic data to train an AI model for better cell segmentation. The researchers found that providing a more diverse dataset during training improves performance.
DynGAN detects and resolves mode collapse by establishing thresholds on discriminator outputs and training dynamic conditional generative models. This improves the diversity of generated samples, surpassing existing GANs and their variants.
Contraceptive services experienced downward trends from an initial increase in the month after Dobbs v. Jackson, indicating growing challenges for access. The study found decreasing workforce numbers providing contraception methods.
Mayo Clinic researchers have invented a new class of artificial intelligence (AI) algorithms called hypothesis-driven AI, which can help discover the complex causes of diseases like cancer and improve treatment strategies. This emerging class of AI offers an innovative way to use massive datasets to guide individualized medicine.
A new dataset has been released that combines molecular information about the poplar tree microbiome with ecosystem-level processes. The dataset provides detailed information on 27 genetically distinct variants of Populus trichocarpa, a bioenergy crop, and includes data on gene expression, soil chemistry, and microbial diversity.
A new study found that non-medical cannabis use is significantly associated with a 96% decrease in odds of subjective cognitive decline. However, medical and dual-use were not significant in reducing the risk of SCD. The study highlights the potential protective effects of cannabis on cognition, but more longitudinal research is needed...
A new satellite dataset provides unprecedented insights into global plant growth, derived from TROPOMI satellite observations. The Comprehensive Mechanistic Light Response (CMLR) gross primary production (GPP) dataset offers a more accurate measurement of plant productivity on a global scale.
Researchers found that doctors can confidently skip 50% of biopsies by combining MRI-based prostate imaging reporting and data system scores with prostate-specific antigen density testing. The study suggests that this approach can decrease patient harm and healthcare costs associated with unnecessary biopsies.
Researchers from the University of Washington created an AI algorithm to analyze infant poses using limited training data. By leveraging generative AI, they were able to produce high-quality results, enabling parents to monitor their babies' daily activities and detect potential health issues early.
Researchers investigate how different VAE model architectures, latent space configurations, and training datasets impact the performance of generative music models with explainable features. They find that measureVAE has higher reconstruction accuracy but lower musical attribute independence.
A recent study found that most Americans support democratic norms, with 17.2% of Democrats and 21.6% of Republicans supporting at least one norm violation. However, the researchers also discovered a divide between everyday citizens and elected officials who are pushing against democratic governance.
A study in São Paulo's central area found that poor wood condition, sidewalk root constriction, and drastic pruning are major predictors of urban tree failure. The researchers propose guidelines for stakeholders to reduce the number of failures, which average 2,000 per year.
Researchers introduced a novel method to augment in situ root datasets through an improved CycleGAN generator, achieving significant enhancements in speed, accuracy, and stability. The approach also boosts dataset versatility by including diverse culture mediums.
A minimal metadata set (MNMS) is established to enable in vivo data reuse and improve the sharing and reproducibility of research data. The MNMS is designed to contribute to making data from living animals compliant with the FAIR data concept, which emphasizes Easy-to-Find, Accessible, Interoperable, and Reusable data.
Researchers developed a machine-learning model to assess prostate cancer biopsy samples, overcoming limitations of traditional methods. The new model, nnU-Net, provides accurate 3D segmentation of glandular tissue structures, leading to better prognostic analyses and potential improvements in patient outcomes.
DomAda-FruitDet is a domain-adaptive anchor-free fruit detection model that achieves impressive average precision scores of up to 94.0% across various fruit datasets. The model effectively bridges the foreground and background domain gaps, enabling accurate and efficient auto-labeling in smart orchards.
A new study warns that rural pupils in England are facing underachievement in education due to socioeconomic factors. Despite better exam results, rural pupils from disadvantaged homes perform worse than urban pupils from similar backgrounds.
A new study refutes previous research suggesting Spinosaurus was an aquatic pursuit predator, highlighting methodological flaws and low accuracy of phylogenetic flexible discriminant analysis. The researchers provide guidelines for future studies to use with caution when applying this methodology to limited datasets.
A team of Rice University researchers has developed a platform for integrating DNA and RNA data from single-cell sequencing with greater speed and precision. The method, MaCroDNA, relies on a classical algorithm to identify matching pairs of data and outperformed state-of-the-art technologies in accuracy measurements.
A new study finds that biodiversity is becoming more similar through time, but an almost equal number of communities are becoming more distinct. The analysis of 527 datasets shows a balanced trend between homogenisation and differentiation across landscapes.
Researchers developed a novel machine learning-based approach to analyze diffuse reflectance spectroscopy data, achieving higher accuracies and speeds than existing methods. The 'wavelength-independent regressor' model overcomes use-error limitations by incorporating diverse datasets, making it suitable for clinical settings.
Researchers optimized polygenic risk scores using ancestrally diverse genomic data to improve accuracy across diverse populations. The recalibrated tests provided a more accurate assessment of disease risk for individuals with varied ancestral backgrounds.
Researchers found that widely used machine learning tools produce biased results for immunotherapy research, as they rely heavily on datasets from higher-income communities. This can lead to ineffective treatments for lower-income populations. The study highlights the need for accurate and unbiased data in machine learning models.
A new study found that 17% of young mothers in Wales experienced at least one child being taken into care between 2014 and 2021. The study also revealed that many care-experienced mothers experience early motherhood and placement instability, leading to concerns about family separation.
Researchers use machine learning to combine mismatched datasets and reduce variation by over 95%, retaining meaningful differences. The approach has potential to provide deeper understanding of normal metabolism and identify biomarkers for disease.