Research announcements from universities, hospitals and scientific institutions. Explore the stories, people and discoveries behind the headlines.
Dates and source material belong to the original release.
Read our research explainersBrowse the collection · original publication order
University of Washington researchers repurposed earthquake-monitoring fiber optics to measure how tilling damages soil structure. The technique revealed that plowed fields lose their capillary networks, causing rain to pool on the surface rather than soak in, increasing erosion and flood risk.
Read the release
An international team led by ISTA will build high-elevation observatories on four continents, train AI models on their data, and produce the first-ever global simulation of mountain water resources — including cryosphere tipping points that most models ignore.
Read the release A six-country team led by Austria's ISTA will combine field observatories in the Rockies, Andes, Pamir, and Himalaya with AI-accelerated physics models to produce the first worldwide reanalysis of mountain water — and identify tipping points before communities lose supply.
Read the releaseBy engineering how different states of light interact inside a lattice of tiny ring resonators, researchers created multiple protected channels along the same pathway — turning what was a single-lane road for photonic data into a robust multi-lane highway.
Read the releaseA manganese-doped bismuth ferrite thin film grown on standard silicon delivers five times the energy-harvesting efficiency of earlier versions, opening a realistic path to mass-produced, lead-free vibration sensors.
Read the releaseAn analysis of 88.8 million deaths across 653 European regions reveals that the poorest communities face twice the mortality risk from air pollution as wealthy ones, even as renewable energy cuts particulate matter by up to 54%.
Read the releaseAn international team used AI to simulate hundreds of rule sets on a mysterious Roman-era limestone board from the Netherlands, matching wear patterns to identify it as a blocking-style strategy game - pushing the genre's known history back by centuries.
Read the releaseResearchers have used terahertz waves to detect electrical activity inside fully packaged semiconductor devices without physical contact, disassembly, or powering down - a first that could transform how we test and monitor electronic chips.
Read the releaseAn international research team developed a stepwise back-correction algorithm that dynamically adjusts rainfall data in hydrological models, improving accuracy by up to 18% across watersheds in Illinois, Colombia, and Brazil.
Read the releaseNew research published in the Academy of Management journal proposes that effective human-AI collaboration requires 'hybrid cognitive alignment' - a gradual process where people and machines develop shared expectations about roles, capabilities, and trust.
Read the releaseESMT Berlin research shows that salespeople avoid pushing novel products because they dread being unable to answer customer questions - a fear of losing face that cuts sales performance regardless of how strong the product actually is.
Read the releaseUniversity of British Columbia researchers tested 12 turf fields across Metro Vancouver and found that crumb rubber infill made from recycled tires releases 6PPD-quinone - a chemical lethal to coho salmon - into stormwater systems at concentrations exceeding toxic thresholds even from fields over six years old.
Read the releaseResearchers at the Technical University of Munich built a wheeled robot that uses a camera, depth sensing, and a language model to construct a 3D map of a room and reason about where lost objects are most likely to be, finding them roughly 30% more efficiently than unguided search.
Read the releaseAn international team found that nonmagnetic impurities in superconductors shift background quantum states away from Majorana zero modes without disturbing the modes themselves, producing clearer spectral signatures that could accelerate progress toward topological quantum computing.
Read the releaseA new knowledge-guided machine learning model from the University of Minnesota combines hydrology physics with AI pattern recognition to predict floods with greater accuracy than either approach alone, eliminating the labor-intensive manual adjustments that current forecasting systems require.
Read the releaseComputer scientists at NYU's Courant Institute developed a bird-flocking-inspired algorithm that preprocesses long documents before feeding them to AI models, clustering sentences by thematic relevance to reduce noise, repetition, and hallucination in generated summaries.
Read the releaseA University of Waterloo study testing 11 large language models across 44 tasks found that even top commercial AI models achieved only 75% accuracy on structured outputs like JSON and XML, while open-source models managed 65% - raising questions about AI reliability in professional software development.
Read the releaseA study in The Lancet Global Health found that climate action consistent with a 2-degree-Celsius target could prevent 13.5 million premature deaths from air pollution by 2050, but equity-based policies that shift mitigation costs to rich nations inadvertently reduce those health gains in developing countries - unless paired with targeted air pollution controls.
Read the releaseThe Deep Underground Neutrino Experiment will generate datasets so massive that traditional analysis cannot keep up. A first-of-its-kind workshop at Rice University mapped out how AI and machine learning will be integrated into DUNE's computing ecosystem.
Read the releaseResearchers used machine learning to screen 660,000 candidate compositions and identified a BaTiO3-based high-entropy ceramic that achieves an energy storage density of 10.8 J per cubic cm with 86% efficiency - a result enabled by its unusual position between two types of ferroelectric behavior.
Read the releaseSwRI received a US Air Force Academy contract to develop and characterize eddy current inspection methods that can detect cracks in aircraft bolt holes through installed bushings, eliminating the risk of damage from removal.
Read the releaseResearchers at Jeonbuk National University developed a clustering-based machine learning framework that trains one model per group of hydrologically similar stations, enabling accurate water level forecasts even where historical records are sparse.
Read the releaseResearchers at Doshisha University developed DUPGT-CDR, a deep learning framework that separately processes positive and negative user ratings to improve recommendations for new-platform users. Tested on Amazon review data, it reduced prediction errors by up to 20.9% compared to existing cross-domain methods.
Read the releaseInsilica's ToxIndex platform uses AI agents to produce comprehensive toxicological risk assessments in hours instead of months. Endorsed by Johns Hopkins CAAT director Thomas Hartung, the platform integrates 600+ open-source models, in vitro evidence from 60 billion data triples, and regulatory databases to deliver fully traceable safety evaluations.
Read the releaseIndiana has signed House Bill 1266, requiring data science integration across all K-12 subjects and creating a dedicated high school data science track covering AI engineering and computer science. The law follows a statewide summit and aims to replicate Indiana's jump to sixth nationally in reading.
Read the release