Articles tagged with Data Sets
Researchers at the University of Washington found that medical AI models rely on shortcuts, which can lead to diagnostic errors and misdiagnosis of COVID-19. These shortcuts ignore clinically significant indicators and rely on irrelevant factors, causing the model to make unexpected associations.
Researchers developed an AI-based tool called SCMER to identify rare groups of biologically important cells from single-cell datasets. The tool was applied to analyze several published datasets, revealing new genes and proteins involved in tumor development and drug resistance.
A new study finds that cultural collectivism is a strong predictor of mask usage in the US and globally, with high collectivistic cultures encouraging masking. The study analyzed datasets from the US and 29 countries, controlling for other factors to confirm the link between collectivism and mask use.
A recent study by GEOMAR scientists has provided a comprehensive understanding of the northern Chilean subduction zone, shedding light on the relationship between earthquakes and tsunamis. The data set, obtained through a unique deployment of ocean-bottom seismometers, revealed that aftershocks were located both beneath and above the p...
A study found that deep neural networks can accurately predict lung cancer type from CT scans, identifying new associations between genes and imaging features. This approach increases radiologists' confidence in assessing tumor types, informing individualized treatment planning.
A comprehensive study finds that many global groundwater wells are at risk of running dry due to declining water levels. The researchers analyzed construction records and monitoring well data from 40 countries, finding that 6-20% of wells may dry up if water levels continue to decline.
Researchers adapted an algorithm to identify similarities in escort ads, highlighting common parts and indicating potential human trafficking activity. The InfoShield algorithm outperformed other algorithms at identifying trafficking ads with 85% precision, flagging them without false positives.
A new algorithm developed by Joshua Welch and his team enables researchers to analyze large datasets using standard computer memory, greatly speeding up single-cell sequencing research. The technique allows for real-time analysis of millions of cells without reprocessing older data.
The expanded patient-level data will facilitate more efficient clinical trial design, enabling generation of novel drug development tools and solutions. The novel tools and solutions will enable better informed inclusion criteria, endpoint selection, and patient enrichment strategies.
Achuta Kadambi's article in Science explores how medical device physics can perpetuate bias across racial and gender lines. He suggests quantifying sample fairness and recalibrating performance metrics to address these issues.
Artificial neurons help decode cortical signals using a new algorithm that automates feature extraction and interpretation. The neural network architecture is automatically tuned to analyze signals from separate neural populations, providing physiologically meaningful results.
A new algorithm, C2FIV, uses facial motion to verify identities, providing an additional layer of security. With a success rate of over 90% accuracy in its preliminary study, the technology has broader applications beyond smartphone access, including workplace and online banking security.
A systematic review of 62 COVID-19 machine learning models found significant methodological flaws, biases, and lack of reproducibility. The models' training data was often poor quality, with issues such as biased datasets, small sample sizes, and a lack of clinical input.
Researchers at Penn State developed FairGNN, a novel framework that estimates sensitive attributes to reduce bias in graph neural networks. The model maintains high performance on node classification using limited user-supplied information while reducing bias.
A University at Buffalo researcher found that the brain's reading network is connected to cognitive domains beyond reading, including math. The study identified a 'connectivity fingerprint' suggesting that reading proficiency affects how we approach tasks and solve problems in other areas.
A graduate student used a class assignment to apply linear algebra techniques to analyze flow field data, identifying dominant modes that capture the most important characteristics of the flow. This technique, called proper orthogonal decomposition, allows for efficient study of unsteady processes with minimal noise and uncertainty.
Colorado State University researchers are part of a NASA team analyzing air quality and health data. The project aims to create a dataset identifying pollutant concentrations across the US from 2006 to present, facilitating studies connecting air quality to health outcomes.
Professor Sun's research focuses on deep learning and meta-learning for recognizing images and videos. Her team is working on a food app that uses AI to track nutrition and achieve a healthy diet, but faces challenges due to cultural diversity.
A new software tool called scfind allows researchers to quickly query datasets generated from single-cell sequencing, identifying which cell types any combination of genes are active in. This enables swift analysis of multiple datasets containing millions of cells by a wide range of users.
A new dataset from ATR Brain Information Communication Research Laboratory Group provides insights into Decoded Neurofeedback, a technique that uses AI and brain scanning to modify fear memories and boost confidence. The dataset includes five studies and could lead to new treatments for PTSD, phobias, and anxiety disorders.
Researchers used medium-resolution satellite images to generate a burned area dataset for sub-Saharan Africa, resulting in significantly higher estimates of burned areas compared to coarse-resolution images. The study estimated fire carbon emissions at 1.44 PgC, substantially surpassing previous estimates.
Researchers found that radar overestimates precipitation rates when partially frozen droplets are larger than their solid and liquid counterparts. The phenomenon, known as reflectivity maxima above freezing (RMAF), is more common on windward slopes of mountain ranges.
Researchers found that neuron structures differ between brain areas and individuals, with variations in shape and curvature affecting cognitive functions. The study used nanotomography to analyze brain tissue samples from 34 schizophrenia and 4 control cases.
Regional disparities exist in responses of secondary pollutants to emission reductions, especially fine particulate matter and ozone. Ozone shows rising signals in most areas, but declining signal exists in North America despite reduced nitrogen oxides.
A new study found that the revised concussion guidelines led to a significant reduction in symptom duration among young athletes. The adoption of active rest, recognizing pre-existing conditions, and educating athletes about recovery improved outcomes, with male athletes experiencing a median duration drop from 11 to 5 days and female ...
A newly merged global surface temperature dataset, including reconstructed land and marine measurements, reveals a consistent increased warming trend compared to previous estimations. The study provides evidence that the globe has warmed at a significantly faster rate than previously thought, with improved coverage of the Earth's surface.
Researchers used 'Neuropixels' probes to capture electrical signals from hundreds of neurons, revealing how visual information flows across the brain. The study found a hierarchy of neural activity, with lower areas representing simpler concepts and higher levels capturing complex ideas.
Computational materials scientists at Ames Laboratory have created an algorithm that uses a hybrid approach inspired by cuckoo birds' nesting habits to find novel high-entropy alloys. The new method significantly reduces the search time for these materials, which are highly sought after for their unique properties and applications.
Researchers at UCLA have developed Diffractive Deep Neural Networks (D2NNs) for all-optical object classification, achieving higher accuracy than individual constituent D2NNs and digital AI models. The success of the ensemble learning approach demonstrates the power of combining multiple predictions to obtain a more accurate prediction.
A new study by Dr Andrea Baronchelli and colleagues reveals a connection between cryptocurrency coding and market behavior. The researchers found that 4% of developers contribute to the code of two or more cryptocurrencies, questioning the transparency surrounding the coding process.
A comprehensive study reveals that Germany's plant diversity has been declining steadily over the past decade, affecting both archaeophytes and neophytes. The decline in species diversity averages around two percent per decade, with many common species being affected.
A team of researchers has found evidence that UV radiation from sunlight significantly affects COVID-19 transmission. By analyzing daily COVID-19 cases and environmental conditions, they discovered a correlation between UV exposure and the growth rate of new cases.
Researchers provide a 4D image of an active linkage zone between two major faults, offering insights into fault behavior and implications for seismic hazard assessment. The study highlights the importance of reevaluating 'silent' seismogenic sources in assessing earthquake risk.
A Purdue University research team uncovered flaws in a prominent dataset used to study brain activity and decode visual perception. The findings challenge previous conclusions on the possibility of reading minds through electric brain signals.
A new method, nnU-Net, has been developed to configure self-learning algorithms for a large number of different imaging datasets, enabling the interpretation of three-dimensional imaging data and distinguishing between tumor and non-tumor tissue.
An international team of researchers has connected specific genetic signals with specific areas of the face, identifying 203 genomic regions that play a role in human facial development. The study sheds light on craniofacial malformations such as cleft lip and palate, as well as evolutionary differences between various populations.
The Leipzig Catalogue of Vascular Plants (LCVP) is the largest and most comprehensive list of scientific names for all known plant species. With over 1.3 million entries, it vastly expands global knowledge of plant diversity and provides a reliable reference for researchers worldwide.
Researchers analyzed a massive dataset of seasonal events across the former Soviet Union and found that species' activities fail to keep up with their cues due to past evolution. Local adaptation patterns result in a massive imprint on nature's calendar, making geographic variation in seasonal timing more pronounced in spring and less ...
A new paper offers a framework for studying online hate and counter speech, analyzing millions of Twitter interactions. Researchers found that organized movements are more effective than individuals in countering hate speech.
A new USC study reveals that AI models lack common sense to generate plausible sentences, despite advances in natural language processing. The research challenges the effectiveness of current benchmark tests and finds that even the strongest models can make silly mistakes.
A new disaster database, DesignSafe, is changing how planners prepare for and respond to natural disasters. The database features over 293 published data sets, including insights on earthquake impact, hurricane damage, and more.
Georgia State University researchers have developed a software tool called COINSTAC, which allows for local data analysis and sharing of results, protecting patient privacy and anonymity. The platform is designed to be compatible with deep learning models, enabling the training of analyses on decentralized databases.
Researchers developed a new method to optimize materials exhibiting metal-insulator transitions (MIT) using Bayesian optimization and latent-variable Gaussian processes. The approach identified 12 previously unidentified MIT materials with optimal functionality and synthesizability.
Researchers establish objective standards for defining and identifying binary tropical cyclones (BTCs), which can bring extreme precipitation and cause serious disasters. The study analyzed two best-track datasets and provides a main standard for defining BTCs, based on separation distance and coexistence time.
A new open-source dataset has been created to provide county-level exposure numbers for tropical cyclones and human health. This tool will enable scientists to analyze multiple storms in different places and time periods, drilling down to see what happens in different health outcomes for people.
For the first time, astronomers have surveyed over 250 million stars in the Milky Way's bulge, measuring their chemical composition and gaining new insights into the galaxy's formation. The data will help scientists understand how the Milky Way formed its central bulge and gain a better understanding of other galaxies.
Recent research shows that chimpanzees, like humans, increasingly prioritize mutual and equitable friendships with others as they get older. Younger adults tend to form lopsided relationships, while older chimpanzees focus on maintaining smaller, fulfilling networks of close friends.
A new data set provides high-resolution daily temperatures from around the globe to study human health impacts from heat waves, risks to agriculture, droughts, and food insecurity. The CHIRTS-daily dataset offers accurate estimates of air temperatures for 1983-2016, supporting efforts to monitor, understand, and mitigate climate hazards.
Researchers have developed machine learning algorithms to predict which RNA-based toehold switches function well, enabling the identification and optimization of these tools. The algorithms analyzed a massive dataset of over 100,000 toehold switch sequences and predicted their behavior with high accuracy.
A novel method for automated pollen analysis has been developed by combining imaging flow cytometry with deep learning, allowing for accurate species identification and quantitative findings in just 20 minutes. The new tool was tested on 35 plant species and achieved an accuracy rate of 96%, outperforming traditional microscopy methods.
Researchers at Princeton University developed a tool to uncover potential biases in visual data sets, such as stereotypical images and underrepresentation. The tool, REVISE, uses statistical methods to inspect data sets for object-based, gender-based, and geography-based biases.
A global study maps areas with high carbon returns from natural forest regrowth, highlighting climate's role in carbon storage. Climate change mitigation strategy: restoring degraded woody vegetation could store substantial amounts of CO2.
Researchers at Mount Sinai Hospital developed a COVID-19 mortality prediction model based on patient's age, minimum oxygen saturation, and type of encounter. The model showed high accuracy (AUC=0·91) in predicting mortality among patients with COVID-19, offering potential for improved prognostication and management.
Regional differences in Russia's alcohol consumption patterns reveal a complex relationship between government measures and unlicensed alcohol sales. Studies have shown that regions with stricter policies exhibit reduced registered alcohol consumption, but unregistered alcohol remains a significant issue.
Researchers developed a machine learning model that analyzes molecular structure to predict enthalpy of formation with better accuracy than traditional approaches. The model's accuracy improves with more data, enabling the development of fully automated algorithms for predicting complex chemical phenomena.
A machine learning study conducted at Princeton University found that the meaning of words is shaped by culture, history, and geography. The researchers analyzed over 1,000 words in 41 languages and discovered that many everyday words have varying meanings across cultures.
Carnegie Mellon researchers create large dataset capturing interaction between sound, action, and vision to improve robotic perception. The study found that sounds can help robots differentiate between objects and predict physical properties of new objects.
A new study published in the Journal of Transport and Health found that walking to work can significantly improve a person's self-rated health compared to casual strolls. People who walked primarily for utilitarian purposes, such as commuting to work, reported better health than those who walked mostly for leisure.
A new study by Cornell researchers uses machine learning to assess the effectiveness of mathematical tools in predicting financial markets. The model can also predict future market movements, a task considered extraordinarily difficult due to markets' massive amounts of information and high volatility.
A new study by scientists at the University of Cambridge's Autism Research Centre found that transgender and gender-diverse adults are three to six times more likely to be diagnosed as autistic. The research used data from over 600,000 adult individuals and confirmed previous smaller scale studies.