Researchers developed machine-learning models to predict mortality in patients after a type of heart attack, known as STEMI. The models, called PERISCOPE, outperformed existing scores in predicting in-hospital and six-month mortality rates.
Researchers introduced a framework for evaluating AI tools in psychological assessment, finding that AI models can score well on accuracy measures while missing key signs of mental health concerns. The study recommends a more rigorous testing process to ensure valid psychological assessments.
ISTA PhD students Riccardo Cadei, Jiale Chen, and Nikita Kalinin have been awarded Google PhD Fellowships for their groundbreaking AI research. Cadei and Chen will receive up to $120,000 in funding to explore cause-and-effect relationships and develop more efficient large language models, respectively.
A new AI system, PI-GAE, predicts aerodynamic performance of shapes, simplifying the design process for engineers. The system accurately reconstructs shapes and predicts performance, with coefficients of determination ranging from 0.946 to 0.992 across four tests.
MIT scientists developed a training method to generate models that simulate climate effects much more quickly than other approaches, allowing for more reliable estimates and unexpected outcomes identification. The technique boosts performance and adaptability while using only a fraction of the data required, significantly cutting produ...
A Saitama University research team developed an architecture called layered attention-enhanced reservoir computing (LAERC) to achieve competitive predictive performance in language modeling while reducing computational burden. The study demonstrates that reservoir-based architectures can provide a practical alternative to traditional t...
A new study using AI algorithms found that climate conditions play a critical role in nitrate pollution in rivers, suggesting that dry climates can elevate nitrate concentrations even in areas without urban or agricultural development. The research, published in Science, reveals that nitrate pollution is a broader problem directly rela...
Researchers at Graz University of Technology developed CLARA, an AI toolbox that accelerates bioreactor simulations by reducing complexity while maintaining physical accuracy. The tool enables precise control of large-scale industrial plants, optimizing processes in fermentation, biomass energy, and polymer production.
Researchers tested 12 AI models and found that pharmaceutical advertisements could shift AI recommendations, making advertised drugs more likely to be selected. The study highlights the need for clear disclosure and separation of advertising from clinical advice in AI-generated health information.
A University of Kansas expert has formed an international group to create conditions for educational revolution with students taking charge of their own learning in the age of AI. The 'Courageous Minority' movement encourages leaders to give teachers freedom to try new things without strict control.
A newly developed AI agent called MLP Copilot has made atomic modeling easier by bridging computational tools and scientific evidence with human supervision. The framework has been used to investigate aluminum nanoparticle oxidation, revealing two behaviors of particles with an existing oxide shell.
Researchers used AI to predict cavity risk by analyzing social determinants, enabling targeted public health programs. Wealth, household density, and water source were strong predictors of cavity risk, with non-White populations and private dental insurance showing protective effects.
A 60-page review assesses the validity of AI-driven science, highlighting three levels of evidence and four constraints: data quality, model interpretability, generative physical hallucinations, and human oversight. The review suggests that the integration of data, compute, and people is key to achieving advantages in AI-for-science.
Researchers have developed a unified AI architecture, WorldParticle, that can simulate multiple physical phenomena, from sliding sand to folding proteins, with one unified artificial intelligence architecture. The system represents every material as tiny particles and predicts where those particles will move one step at a time, adaptin...
A new study proposes a holistic theory of AI fairness and an applicable framework for organizations to manage the complicated issue of fairness in AI-automated decision making. The proposed design theory, FAIR, addresses conflicting concerns through regular human-AI collaboration and treating AI fairness as a continuously managed probl...
Researchers at Lehigh University are developing a model that uses causal auditing and symbolic reasoning to improve the transparency and accountability of LLM-based recommender systems. This could help users better understand why a particular recommendation was made and what information influenced it.
AlphaOracle cuts analysis time by 64% and 78.7% of experts rate its usefulness; the system uses four stages to trace characters, including morphological analysis and contextual alignment, and is presented as a step for partially documented scripts
The National MS Society is launching a funding opportunity to support innovative screening technologies for early MS detection. The goal is to accelerate progress in identifying individuals at risk of MS, ultimately reducing time to diagnosis and improving long-term outcomes.
Researchers developed a mobility-network framework that estimates city-level CO2 emissions with R2 up to 0.99 across China, Italy, the US, and Mexico. The framework used anonymized cellular signaling data and machine-learning models to predict emissions in near real-time, providing a concrete solution for climate policymakers.
Experts from Hackensack Meridian Health and other institutions emphasize the need to make human-AI interactions visible to clinicians to advance patient care. The authors propose using divergence visibility to reconcile human and machine opinions, mitigating liability and improving clinical decision-making.
A machine learning study decoded biochar's reactive radicals from its elemental composition, identifying hydrogen-to-carbon ratio and oxygen content as leading predictors of radical concentration. The findings provide a practical starting point for matching biochar chemistry to use in environmental applications.
Foundation models in Earth system science have five barriers and six paths to overcome, including a unified Earth representation engine and physics-embedded neural architectures. The authors propose six paths to models that respect physics and invite scrutiny, with a focus on data, metadata standards, and calibrated uncertainty.
Researchers at the University of Sydney have identified a fundamental mathematical reason why AI systems can learn different behaviors even when trained under almost identical conditions. The study found that extremely small differences in a model's starting conditions can lead to different outcomes, revealing a fractal structure in the
Researchers developed StereoGeoBox, a system that estimates fish dimensions and weight without generating a complete depth map. The system achieved the lowest estimation errors compared to conventional methods, while producing fewer outliers and demonstrating real-time performance.
Researchers develop Physics-based Learning for Imaging Planets (FLIP) to remove starlight from telescope data, revealing exoplanets previously hidden. The method improves detection sensitivity, enabling scientists to study more planets and uncover previously missed worlds.
Researchers develop TACToC, a communication system that transmits only the information required for AI tasks, reducing communication costs by up to 14.5×. mzCache reduces AI time to first response by up to 5.5×, shortening response delays for smartphone AI.
A new system called UQ-DAAAM identifies and corrects unreliable object descriptions in a robot's memory, improving its ability to interact with its environment. This method boosts the robot's confidence and improves the accuracy of its answers to questions about objects, their locations, and when it saw them.
Researchers built a machine learning model using 14-year data from 939 SSHL patients to predict healthy ear incidence. The model identified age, hearing loss severity, gender, and cardiovascular and inflammatory indices as key predictors.
The Seoul National University research team, led by Professor Kyomin Jung, has achieved significant milestones in AI research, including 44 papers accepted to top conferences in two years. Key findings include advancements in AI reasoning, multimodal AI, agent safety, and language generation, demonstrating the team's expertise in natur...
A research team led by Arslan Munir evaluated quantum machine learning approaches for predicting lung cancer using limited symptom and risk-factor data. D-Wave's annealing quantum computers achieved the highest accuracy, reaching 94.31% in simulation and 91.90% on quantum hardware.
Researchers developed a technology called 'CluSTER' that selects and distributes training data, reducing repetitive learning and training time by up to 69.6%. This enables faster development of AI tailored to specific tasks, such as corporate customer service or medical information retrieval.
Researchers developed an automated platform to track perovskite film formation in real-time and use machine learning to improve solar cell performance. The platform enabled the identification of optimal conditions for droplet behavior, leading to a power conversion efficiency of 24.4% in a lead-based perovskite solar-cell device.
Causal machine learning and reinforcement learning can bridge the gap between predictive models and personalized therapy in critical care. Causal ML estimates individualized treatment effects, while RL generates adaptive treatment policies over time. Future directions include reliable treatment effect estimates, conservative Q-learning...
NTU Singapore and Google are expanding their partnership to accelerate AI adoption across education, research, and workforce readiness. The collaboration will support faculty training, professional certification, and region-specific AI initiatives, while advancing frontier scientific research and AI safety.
Researchers developed a machine learning model using routine non-invasive prenatal testing data to predict placenta accreta spectrum disorders. The 15-gene multiclass 4P-model achieved high accuracy in differentiating controls, accreta, and increta/percreta, expanding the diagnostic window into early pregnancy.
Penn's Fast Feature Field transforms event-camera data into information that many different kinds of robots can use, improving robots' ability to recognize objects, estimate distances and perceive motion. The tool could transform robotics by allowing for the widespread use of event cameras in self-driving cars and drones.
Researchers propose using psychology's existing protocols to assess LLMs' intelligence, biases, and social skills, but warn of potential pitfalls in understanding AI capabilities
Researchers at Tohoku University propose 'interfacial intelligence' framework to design polymer brushes with improved predictability and reliability. The approach combines polymer physics, machine learning, and simulations to understand how polymer brushes interact with their surroundings.
The team's research on reducing environmental noise in AI and generating diverse synthetic training data for large language models has been accepted to top AI conferences. Their methods have been shown to improve AI performance and efficiency across various fields, including robotics, reinforcement learning, and quantum AI.
Researchers developed a diagnostic signature for sepsis using machine learning and single-cell RNA sequencing, identifying four genes associated with immune dysregulation and organ failure. The signature may help support individualized risk assessment and treatment optimization.
Researchers use AI to enable spectrometer to learn best pulse sequence directly from sample, improving signal detection in NMR analysis. This approach shows promise for developing better magnetic-resonance-based quantum sensors and advanced laboratory applications.
Researchers developed a machine learning model to estimate leaf number and determine optimal fruit load in pear trees using 3D point cloud analysis. The model achieved high accuracy and stability, and field validation showed improved fruit quality and yield, enabling timely thinning decisions.
Researchers will analyze data from over 320,000 patients to develop an AI model that predicts which patients may develop acute kidney injury after surgery. The goal is to tailor prevention strategies to individual patients, reducing the risk of chronic kidney disease and death.
A novel framework uses AI and social media to identify promising brand collaborations, providing practical guidance for brand managers. The framework has been tested and validated, showing a nearly 90% chance of correctly identifying favorable brand pairs.
Researchers use explainable AI to uncover patterns hidden in centuries-old paintings, providing a new tool for investigating painting authenticity. The AI framework identifies statistical patterns across characteristics like edges, color, and brushstrokes to produce a score estimating how strongly a painting aligns with Rembrandt's work.
A new AI model, developed by the University of Cincinnati Cancer Center, aims to improve breast cancer survival prediction by combining statistical methods and deep learning. The model analyzes clinical and genetic data to provide more personalized care decisions for each patient.
The project aims to develop shared cyberinfrastructure that enables realistic, reproducible experiments for AI-enabled robots and cyber-physical systems. The framework will provide research-grade workflows, digital twin environments, and experimental examples to support hands-on training and mentored research.
Researchers found that combining Global Navigation Satellite System (GNSS) observations and modern bias products can deliver near-instantaneous centimetre-level positioning. Machine-learning attribution frameworks identified key factors driving remaining errors, providing a data-driven complement to conventional GNSS diagnostics.
A new AI tool, circMAC, predicts microRNA binding sites on circular RNAs at nucleotide resolution, providing more focused candidate regions for experimental validation. The tool outperforms existing RNA language models and demonstrates its potential in supporting precise studies of RNA-mediated regulation.
Researchers have developed a metamaterial that combines stiffness, electromagnetic absorption, and noise reduction, inspired by the intricate scales of lepidopteran insects. The material, just 24mm thick, can support heavy loads, absorb microwave signals, and silence low-frequency noise, solving a long-standing trade-off in flight mate...
A new materials library was created to improve sodium-ion battery performance, with iron and vanadium linked to higher energy storage and stability. The library could eventually allow AI to identify new battery compositions before they are made, accelerating the development of better sodium-ion batteries.
A Lehigh-led team is developing an AI platform called UniBrain to simplify brain imaging analysis, reducing the time and expertise required. The platform aims to unify image preprocessing and downstream brain network analysis, integrating information from multiple imaging modalities.
The New Jersey Supercomputer Initiative (NJSCI) aims to advance research, innovation, education, and workforce development across New Jersey. The initiative will establish a shared supercomputer to support researchers, students, and businesses, enabling them to tackle complex challenges like cancer research and flood risk mapping.
Researchers at Chalmers University of Technology developed a new optimisation method for near-optimal large-scale fleet control, accounting for uncertainties in real-world transport systems. The method combines machine learning techniques with advanced optimisation models to predict travel demand and charger availability, improving fle...
Researchers developed a machine learning model that predicts functional decline after heart failure, identifying patients at risk of losing independence despite surviving hospitalization. The model outperformed traditional prognostic approaches, highlighting the importance of physical function at discharge.
Researchers investigated the association between Alzheimer-related changes and disease progression in ALS, finding that coexisting changes may affect prognosis. The study suggests that biomarkers like amyloid beta and NfL could improve diagnosis and prognosis, and may contribute to better patient stratification and more efficient clini...
SwRI's automated survey uses machine vision software to identify unrestrained dogs in San Antonio, helping the city better characterize their population. The project aims to promote responsible pet care and strengthen community connection.
Researchers developed a machine learning approach to reconstruct missing precipitation patterns in global rainfall maps, using a 3D U-Net and 3D condition encoder to capture space and time variations. The model achieved high prediction accuracy and produced consistent rainfall patterns across locations and time.
Researchers introduce an integrated framework linking raw materials, manufacturing processes, and real-world applications to design more reliable and sustainable carbon technologies. The review highlights the importance of understanding biomass composition and processing methods to produce high-performing carbon nanomaterials.
A team at Kyushu University analyzed 1,623 annual securities reports from 340 Japanese food and beverage companies, finding that firms place greater emphasis on social issues, while environmental disclosure remains relatively limited. Disclosure priorities shifted after 2022, with greenhouse gas emissions showing the sharpest increase.