Articles tagged with Data Sets
The All of Us Research Program has made strong progress in advancing precision medicine by collecting diverse participant data. The program aims to include 1 million or more people from diverse communities, with over 230,000 participants enrolled and 80% from underrepresented groups.
A new analysis of state, county, and federal data shows a significant decline in criminal sentencing racial gaps since the mid-1990s. The overall sentence length gap between whites and blacks in federal courts decreased by more than 80% between 1996 and 2016.
The project aims to advance research by creating a resource for researchers around the globe to access data on craniofacial development and dysmorphologies. Craniofacial conditions affect approximately 600,000 people in the US, with cleft lip and palate being amongst the most common birth defects.
Researchers from the University of Maryland developed a novel approach to generate questions that challenge AI language models. By collaborating with humans and computers, they created a dataset of over 1,200 computer-stumping questions that revealed six different language phenomena consistently stumping computers. This work provides i...
A recent study published in The Lancet found an AI model can identify patients with intermittent atrial fibrillation using a 10-second ECG test. The technology achieved an overall accuracy of 83% and has the potential to aid doctors investigating unexplained strokes or heart failure.
Researchers demonstrate that even anonymized datasets can be traced back to individuals using machine learning. The study shows that allowing data to be used for AI training while preserving people's privacy requires more than de-identification techniques.
Researchers at Rush University Medical Center have developed a new composite ranking system that aggregates data from five leading hospital rating systems into a single score. This system aims to provide consumers with a clear and easily understandable measure of hospital quality, helping them make informed decisions about their care.
Researchers developed an AI-based system, SpoilerNet, to flag spoilers in online book and TV show reviews. The tool achieved 89-92% accuracy in detecting spoilers, with most errors caused by distracting words.
A new study using satellite data reveals large inequalities in progress towards universal access to electricity in sub-Saharan Africa. The research provides a high-resolution gridded dataset to assess electrification in the region, enabling policymakers to prioritize support for infrastructure expansion and track progress over time.
The neuroimaging community has made significant strides in collecting large-scale datasets, but many contain insufficient data for reliable measures of brain connectivity. Increasing reliability can dramatically reduce sample size requirements, improving the field's ability to detect effects and address reproducibility concerns.
Researchers developed a system called Northstar that runs on touchscreens and enables users to manipulate datasets, uncover trends and patterns. The new component VDS instantly generates machine-learning models to run prediction tasks, democratizing data science and making it accessible to everyone.
A new system uses a single labeled scan and unlabeled scans to automatically generate a large dataset of distinct training examples. The system's approach learns anatomical, brightness, and contrast variations from unlabeled scans to synthesize realistic and accurately labeled scans.
A UCLA-led research team developed FEAST, a faster and more accurate tool to determine microbiome origins. The software can analyze large genetic information datasets in just a few hours, offering insights into the unseen ecological processes affecting human health.
A team of researchers from the University of Illinois has developed a new technique to predict high-yielding crop traits by stacking six machine learning algorithms, improving predictive power by up to 15 percent. The approach empowers scientists to leverage computational analysis to translate their datasets into beneficial results.
Engineers have released a freely available quality-controlled dataset from 1,287 residential installations across Australia to improve solar forecasting and grid management. The dataset provides six months' worth of measurements from three different cities, allowing for controlled analysis and spatial analysis.
A study of English secondary school placement systems found that parents in areas with more limited choices tend to make less ambitious school selections. Despite making more choices, students from minority backgrounds have lower chances of receiving an offer from their first-choice school.
Researchers have created a new dataset called BOLD5000, which comprises brain scans of four volunteers viewing 5,000 images. This dataset allows cognitive neuroscientists to leverage deep learning models that have improved artificial vision systems.
Shambhala helps process and compare RNA-Seq data from different equipment, allowing for better study of gene expression and its relation to diseases. The tool facilitates harmonization of large datasets, enabling comparisons across various samples and biological organs.
MIT researchers develop a method to extract comprehensive samples of massive cell datasets, preserving rare cell types and their biological information. The 'sketching' approach generates compact summaries of large datasets in minutes, making it practical for biological studies.
Research finds that current affiliation's prestige shapes the scientific productivity of early-career faculty, not past training environment. Faculty at more prestigious institutions produce significantly more scholarly output than those at less prestigious institutions.
Researchers expand earthquake catalog by a factor of 10, revealing daily quakes at 495 locations across the region. The study's template matching technique identifies tiny tremors, expanding our understanding of seismic events and their impact on the region.
The ML-DSP tool can provide conclusive answers to complex questions about species, DNA, and viruses. It has been shown to outperform other classification software tools in terms of accuracy and speed.
Polygenic scores developed using European data are more accurate for people of European ancestry than those of African or East Asian ancestry. Researchers recommend increasing minority population representation and making data accessible to improve healthcare outcomes.
Researchers from Lancaster University found that April Fools hoaxes share similarities with fake news stories, including simpler language, longer sentences, and less formal tone. The study provides new insights into detecting disinformation and may aid in identifying 'fake news' articles.
A new deep-mapping computer model detects visual changes to individual properties, allowing researchers to track gentrification in neighborhoods and cities. The model achieved 95% accuracy in detecting gentrification indicators, revealing detailed maps of spatial evolution of gentrification through time.
Researchers analyzed fossil pollen and sedimentary records to study Holocene climate variability's impact on the northeastern US. The findings suggest rapid ecological changes in response to changing climate conditions.
Researchers found that evacuees who returned to their homes after initial evacuation had more children, while those who stayed in western Finland had fewer. Those who intermarried with the host population increased their social status, but were less likely to have children.
Researchers can create synthetic networks that satisfy differential privacy requirements, enabling broader sharing of confidential data for scientific discovery. These altered datasets capture most statistical features from the original network, making them suitable for various researchers.
A new method called FIt-SNE has been developed to speed up the analysis of single-cell gene expression data, reducing rendering time from over three hours to just fifteen minutes. This innovation allows researchers to capture rare cell populations and visualize thousands of genes at the level of single cells simultaneously.
A study suggests that empty homes tax could generate income for local governments and reduce demand from foreign investors to increase housing affordability. Low-use properties are concentrated in desirable areas, where building more homes may not solve the problem.
A new study found Hispanic males are twice as likely to have fatal interactions with the police in areas with high Hispanic populations and agencies with more Hispanic officers. The research suggests reforms within police ranks are needed, even in diverse forces.
Researchers developed a deep-learning model that detects fake news by analyzing language patterns, finding favoritism towards exaggerations in false stories. The model achieved high accuracy in distinguishing fact from fiction, particularly when tested on novel topics.
Researchers developed a new method to identify significant ties in social networks, controlling for individual activity levels. This approach accurately mimics real-life school class clusters and has applications in various complex networks, including interbank markets and face-to-face interactions.
A novel valuation method reveals that stock options granted to women and senior managers are worth more because they hold onto them longer. Options with less frequent vesting dates also cost companies significantly more.
A new AI model has been developed to translate mouse model data to human disease conditions, increasing the accuracy of extrapolation by up to 50%. The model, known as Found In Translation (FIT), has been tested on 28 different human diseases and uncovered novel disease-associated genes.
A new study using crowdsourced field size data shows that smallholder farms contribute substantially to world food production, making up 40% of the global agricultural area. The research provides a detailed map and estimates of field sizes at the global, continental, and national levels.
Researchers identify potential causes of bias in machine learning systems and demonstrate how changing data collection methods can reduce bias without compromising accuracy. They suggest identifying clusters of patients with high disparities in accuracy to inform data collection decisions.
Researchers found that small ParkScore improvements lead to increased physical exercise among residents, with cities like Minneapolis and San Francisco ranking high. The study suggests that investing in park infrastructure can positively impact public health, with a collective effect of improved fitness and well-being.
A study by MIT researchers found that about one-third of Requests for Comments on Wikipedia go unresolved, citing factors such as poor initial arguments and excessive discussion. The researchers developed a machine-learning model to predict when RfCs may go stale, and recommended digital tools to improve deliberation and resolution.
A recent study from the University of Colorado Boulder found that academic ideas tend to spread faster from highly-ranked universities, even when the quality of the idea is mediocre. This suggests a power imbalance in academia, where elite institutions have an advantage in spreading their research to others.
Researchers at KAUST have developed VR apps to visualize and interpret complex datasets in three dimensions. These apps use immersive technologies to allow users to interactively explore features of the data, leading to more accurate statistical models.
A new paper highlights the importance of open and usable data sets in ecology and other field research. By sharing data and code, scientists can save time and effort, as well as foster critical inquiry and new knowledge.
A team of researchers has developed a mutual information approach to interpreting atmospheric data collected over an 18-year period, finding strong correlations between new-particle formation and water content, sulfuric acid concentration, temperature, and relative humidity.
A study published in ACS NSQIP demonstrates a sustained benefit on colorectal surgical outcomes, with significant reductions in complications such as surgical site infections and sepsis. The program's introduction of procedure-targeted datasets also led to improved outcomes, including increased early discharges.
Researchers at Dartmouth College used the Bible to develop an algorithm that can convert written works into different styles for different audiences. The study, published in Royal Society Open Science, trained on over 1.5 million unique pairings of source and target verses from various versions of the sacred texts.
A study led by OHSU has published the largest cancer dataset of its kind, involving hundreds of patient samples and revealing new insights into acute myeloid leukemia (AML). The dataset may help researchers and physicians solve specific questions about AML treatment options.
Two types of flash drought have been identified in China, Type I driven by high temperatures and evapotranspiration, while Type II is initiated by rainfall deficiency. Both types have increased significantly over the past 30 years, with a two-fold greater increase for Type I.
A deep learning approach can identify nuanced mammographic imaging features specific to recalled but benign (false-positive) mammograms, distinguishing them from those identified as malignant or negative. The study achieved an area under the curve (AUC) of 0.76-0.91, indicating high performance in detecting false recalls.
Researchers at the University of Sussex have created a large dataset that can be used to develop smartphone apps that detect transportation modes, predict road conditions, and offer route recommendations. The dataset has achieved high accuracy rates in recognizing modes of transport, paving the way for innovative mobile applications
The new global marine environmental forecasting system will provide accurate predictions of ocean temperature, salinity, velocity, and tidal currents with a horizontal resolution of 10 km. The system's accuracy is expected to reach an advanced international level, enabling clear identification of ocean mesoscale phenomena.
A new MIT model has been developed to reduce false positives in credit card fraud detection, achieving a 54% reduction in incorrect flagging of legitimate transactions. By extracting more than 200 detailed features for each individual transaction, the model can better pinpoint unusual spending habits and improve accuracy.
Researchers at Northwestern University have identified four clusters of personality types using comprehensive data from over 1.5 million questionnaire respondents. The findings challenge existing paradigms in psychology and provide new insights into human behavior.
Researchers at MIT develop Temporal Relation Network (TRN) module to help CNNs recognize activities by observing key frames. The module achieves top accuracy of 95% in activity recognition on Jester dataset, outperforming existing models.
Researchers found that overconfident CEOs are 33% more likely to be sued by shareholders. A shareholder lawsuit can curb future risk-taking behavior, leading to lower confidence and more prudent actions. This study highlights the importance of shareholder power in regulating CEO behavior.
A Lehigh research team is developing a 'Google for research data' that can assist scientists in locating datasets across various disciplines. The team aims to create a domain-agnostic search engine that utilizes user-centric methods to develop dataset search tools and novel indexing techniques.
Researchers from the Santa Fe Institute developed a new algorithm called SpringRank that analyzes wins and losses in networks to predict outcomes. The algorithm outperformed others in efficiency and accuracy, even when applied to diverse datasets such as NCAA basketball teams and animal social behaviors.
The DeepLesion database is the largest publicly available medical image dataset, containing over 32,000 annotated lesions from 10,000 case studies. It has tremendous potential to jump-start the field of computer-aided detection and diagnosis.
A Massachusetts-based study found that opioid overdose events decreased during pregnancy but increased in the postpartum period, particularly when fewer resources were available to mothers with substance use disorder. The research team developed a unique dataset linking statewide resources and found that women with opioid use disorder ...
A new approach uses machine learning to generate computer-generated X-rays to supplement real images, increasing the size of training sets for AI systems. This method improves classification accuracy for common and rare conditions by up to 40%, overcoming a challenge in applying artificial intelligence to medicine.
A new study by Queen Mary University of London found that 58% of online video piracy is concentrated in just two locations, making them susceptible to detection by copyright enforcers. The research also revealed a global network of streaming cyberlockers and third-party indexing services facilitating piracy.