Research applying artificial intelligence across science and engineering. Covers machine learning, sensing and quality control, immune and health prediction, and 5G networks.
The review systematically maps out the research landscape of wearable AI doctors, combining power-harvesting approaches with self-powered intelligence. Flexible electronics enable devices to withstand dynamic environments, while on-device AI inference processes data locally for real-time analysis.
Researchers create PhishLumos AI system to detect phishing campaigns by analyzing infrastructure clues, achieving 8-day faster detection than experts. The system uncovered over 190,000 new links, with 92% later flagged as malicious, outperforming content-centric approaches.
A new AI model analyzed 30 years of EHR data to identify patients with primary aldosteronism, a common cause of high blood pressure. The model correctly flagged over 90% of cases while missing fewer than 10%, suggesting it could improve screening for this condition.
The system uses two AI agents to comb through publicly available data and conduct life cycle assessments, achieving an average error rate of 5%-19% similar to expert-led LCAs. The team also developed a new method to bypass detailed data collection and estimate carbon footprints for unknown devices using 'nearest-neighbors' approach.
The study synthesizes recent advances in single-cell and spatial transcriptomics to identify tumor-enriched cell subsets closely related to prognosis and treatment response. The review introduces the
Rockefeller University Press has partnered with Cashmere to enable safe and transparent AI inference use of its journals, protecting authors' work while providing visibility and revenue transparency. The partnership uses Cashmere's infrastructure suite to manage AI-powered research applications.
Researchers at Princeton University used AI to analyze how drugs affect cell structures, finding new shapes linked to disease and discovering a novel drug effect. The neural network identified cap, necklace, and flower shapes, with the latter indicating a previously unknown role of an enzyme in maintaining nucleolar organization.
Researchers developed a new magnetic memory material that can be rewritten using laser light, allowing for faster and more energy-efficient storage and processing of information. This breakthrough could help reduce power consumption in data centers and support future high-speed information systems.
A new AI model has revolutionized molecular simulations, enabling researchers to predict molecular behavior and identify promising drug candidates more quickly. By analyzing over 12,500 organic molecules, the model has demonstrated accuracy and consistency with physical laws.
A team of researchers from The University of Osaka has developed a new approach for depth reconstruction from defocus, estimating distances by analyzing blur in an image. Their method combines a coded-aperture camera with diffusion-model-based AI to accurately estimate depth and produce high-quality images.
Researchers warn that open-source AI could increase environmental pressures, deepen technological inequalities, and spread misinformation. To mitigate these risks, the authors propose four governance actions to ensure AI contributes positively to the Sustainable Development Goals.
The ACM Technology Policy Council's TechBrief examines agentic AI's legal liability, security risks, and workforce impacts. Existing frameworks fall short in addressing accountability questions, highlighting the need for defined authentication and delegation standards, robust audit trails, and sector-specific guidance.
Researchers at the University of Washington are using AI and quantum computing to design new materials with unique properties, such as superconductivity and entanglement. The tools are helping to power the growing field of quantum computing and could lead to breakthroughs in energy-efficient electronics.
Researchers at MSK discovered that tumors with low levels of NF2, a tumor suppressor gene, are less likely to respond to immunotherapy. They also found that combinations of VEGFR-TKIs and immunotherapies can be effective but have limited long-term benefits due to paradoxical effects of hypoxia on treatment response.
Researchers have developed a digital brain twin using multimodal data, providing anatomically accurate brain models that can be used in experimental and clinical settings. The models were reconstructed simultaneously with brain anatomy and dynamics from neural data.
The Luxembourg AI Factory released an open-source tool to accelerate testing for trustworthy artificial intelligence. The AI Assessment Sandbox Configurator enables rigorous AI testing at scale, addressing regulatory requirements and ensuring trustworthiness.
A new study explores how transfer learning reduces computational costs in cosmological simulations while revealing risks of negative transfer, which can hinder learning new physics. Transfer learning can accelerate inference but may also push AI systems toward incorrect interpretations of new effects.
The Acceleration Consortium and Structural Genomics Consortium collaborate to develop new drugs using AI-driven lab capabilities. The partnership aims to speed up the discovery of bioactive molecules, advancing human health and disease understanding.
The new project aims to build self-regulating AI agents that can recognize uncertainty, explain their decisions using past experience, and improve their understanding of the world through corrective feedback. The goal is to make future physical AI systems more trustworthy and useful in real-world settings.
Researchers found that friendly, choice-oriented language in AI chatbots boosts patient engagement, while aggressive phrasing and blurred human-AI boundaries put them off. Patients appreciate personalized interactions, but are wary of data security concerns.
FireANTs, an open-source algorithm, combines AI optimization and geometry to quickly match complex medical images. The new method can accomplish what took weeks in minutes, detecting subtle changes that signal disease or cognitive decline, making it practical for clinical practice.
Teachers tend to accept overly harsh AI grades more than human grades, resulting in a 22% larger gap between their score and the fair score. Teachers are more likely to accept strict AI grades if they view the system as competent and accountable.
A machine learning model, trained on 19 key parameters, can predict wind shear events with a minimum of 15 seconds warning. The model's outputs showed deviations from real outcomes within 5% across all forecast horizons, suggesting improved aviation safety.
A study published in Radiology found that three commercially available AI-based computer-assisted detection (AI-CAD) systems can identify early mammographic signs of breast cancer up to 6 years before diagnosis. This could help radiologists spot potential future cancers and allow for earlier intervention.
TurboLynx, developed by POSTECH researchers, analyzes complex, interconnected data up to 184 times faster than existing systems. The engine groups similar data together and processes them collectively, reducing unnecessary memory usage and enabling efficient analytical queries.
A new ultra-lightweight AI model, Multinex, advances low-light image enhancement by leveraging classical colour vision theory and Retinex principles. The model outperforms comparable compact systems, recovering detail and clarity from previously unusable images.
Researchers at Tohoku University's Advanced Institute for Materials Research have developed a method to summarize decades of scattered literature data into actionable information for catalyst design. By combining human intelligence, regression models, and AI agents, they can uncover new discoveries hidden in the literature data.
A new study found that AI-powered chatbots can make vaccine-hesitant parents more likely to say they will immunize their children against HPV, but no more than standard written public health materials. Additionally, the effects of the chatbots did not last longer than those of government health materials.
Researchers have developed an AI-powered safety switch for future cell therapies using caffeine. The platform, CODS, rapidly separates engineered proteins in living cells and triggers cellular responses on demand.
Participants trusted AI more for large-scale scanning tasks like identifying red flags in social media posts, while trusting humans for nuanced fact-checking requiring piecing together evidence. The findings suggest that effective fact-checking tools should provide accurate results and explain how they're reached.
Researchers at Emory University discovered a flaw in reinforcement learning used to guide sepsis treatment, which can result in either overtreatment or undertreatment. The team developed a simple workaround to avoid the flaw, leading to an 8-10% decrease in patient mortality.
Researchers from UMass Amherst have developed a new AI architecture called ANT that enables continuous learning and reduces energy consumption by orders of magnitude. Unlike human brains, which operate asynchronously, modern deep neural networks rely on synchronized computations, leading to high energy demands.
A neural network-based machine learning model accelerates diffuse optical tomography by over a million-fold, enabling real-time diagnosis. The model accurately reproduces signals even for unseen parameter combinations, with each inference taking approximately 2 milliseconds.
Researchers developed a transistor technology that enables a single device to perform multiple circuit functions simultaneously, simplifying circuit design and increasing data processing speed. The new approach reduces required transistors by 75% and increases data processing speed fourfold.
Researchers developed an interferometric second-harmonic generation imaging approach to identify antiparallel domains and detect hidden structural defects in hBN thin films. The study finds that SHG intensity is closely associated with differences in crystal orientation and destructive interference between domains.
A study found that passive AI use can lead to a decline in outcome satisfaction of 21% compared to manual writing, while collaborative AI use showed similar scores. Passive AI use also reduced feelings of ownership and perceived meaningfulness by nearly 20%.
UT San Antonio's Jeff Prevost is appointed to the Texas Quantum Initiative Advisory Committee, guiding strategic investment and collaboration to advance the state's leadership in quantum research and technology development. The committee aims to develop a strategic plan supporting the growth of Texas' quantum economy.
A University of Bristol team has developed an AI system that can automatically find, name, and follow individual animals in footage, saving thousands of hours for researchers. The system, part of the SA-FARI project, uses a foundational Vision-Language Model to precisely identify objects in images or videos.
Researchers used tissue samples, pathology images, and clinical data from 672 patients to train AI models to classify meningioma subtypes and predict recurrence risk. The findings suggest that AI can help clinicians obtain more detailed tumor information without requiring advanced genetic testing.
A digital 'super-brain' with physics-based knowledge significantly speeds up the design and development of optical components, such as those for quantum computers and camera lenses. By integrating physical principles into machine learning algorithms, researchers reduce simulation time from months to days.
Researchers developed ADASPEC to speed up multilingual AI systems by dynamically adapting to different languages during inference. The framework generates instruction data in any desired language using the target LLM itself, reducing unnecessary vocabulary computations and achieving faster and more stable multilingual inference.
Researchers have developed soft, brain-inspired electronics that can sense, store, and process information while conforming to biological tissues. These devices mimic the chemical processing of the human brain, executing complex tasks like heart rhythm classification at ultra-low voltages.
Researchers developed CardioNVT, an AI platform for high-throughput cardiomyocyte ploidy assessment. The platform accurately recognizes and segments cardiomyocyte nuclei from DAPI-stained images, enabling the analysis of nuclear volume distribution and inferred ploidy.
IMDEA Networks has secured four SNS JU projects to develop next-generation 6G networks in Europe. The projects address key challenges including sustainability, security and industrial transformation. IMDEA Networks is leading two of the projects with principal investigators Marco Fiore and Joerg Widmer.
Boston University has joined the NSF-funded AI Institute for Artificial Intelligence and Fundamental Interactions to unlock new discoveries in physics using AI. The institute, which includes top local universities, aims to develop new approaches to AI by applying physics principles.
A new tool developed by Concordia researchers uses artificial intelligence to plan surgical schedules, reducing wait times and minimizing disruptions. The system can adapt to emergency surgeries while keeping non-emergency cases on schedule.
Researchers from MIT and IBM create a state-of-the-art dataset called ChartNet, which includes over a million varied charts. The dataset is designed to teach vision-language models how to effectively interpret charts, enabling them to outperform commercial models on tasks like data extraction and chart summarization.
A recent study in mice reveals a shared hub of memory cells that links incoming and outgoing signals, allowing the brain to reuse some cells to store many different memories. This 'switchboard' helps maintain stability while learning new information.
Researchers developed a machine learning framework to detect and classify shell-crushing sounds, enabling the quantification of predator impacts on mollusk populations. The system uses multi-step approaches and demonstrates strong performance in controlled tank conditions and field settings.
A new UN report details the environmental costs of artificial intelligence, including its burgeoning electricity use, carbon emissions, water footprint, and land occupation. The investigation finds that AI's expansion involves significant energy consumption, leading to substantial CO2, water, and land footprints.
A new framework, VIBEMed, uses multi-agent collaboration to break complex clinical decisions into specialist roles and employs a three-level self-evolution mechanism to improve performance over time. This approach demonstrates superior performance in complex medical reasoning and treatment planning tasks.
The RBA hydrogels can stretch to more than three times their original length while remaining mechanically stable. They were used as strain sensors to detect subtle facial movements and distinguish between walking, jogging, and running in real time.
A recent EHU study reveals that overreliance on generative AI tools can undermine key learning skills. However, students with strong self-regulation abilities are less likely to fall into this trap, as they tend to use AI as a tool for speed rather than relying solely on its responses.
A novel deep learning framework integrates high-resolution pathology images with spatial transcriptomics and proteomics to reveal complex intra-tumor heterogeneity in colorectal cancer. The approach identifies distinct molecular subtypes that correlate with immune cell infiltration patterns.
A new set of consensus standards for classification, annotation, and quality control is being implemented for artificial intelligence applications in dry eye imaging. These standards aim to enhance consistency and multicenter collaboration in AI-assisted diagnosis.
Researchers at King's College London developed a new algorithm that can automatically explain why some self-driving cars crash. The approach analyzes past events to identify the root cause of failures in complex and rare cases.
Researchers at Viva Bem have developed an AI method that identifies states of anxiety with high accuracy using smartwatch data. The technology aims to provide a layer of proactive monitoring for users, alerting them to recurring anxiety episodes and recommending specialist consultation.
Researchers from Drexel University developed BioCoach, a program using AI and computer vision to analyze video and provide form coaching in real time. The system analyzes visual appearance and motion patterns, as well as 3D skeletal movements and body shape, to deliver detailed biomechanics-based feedback.
A new research initiative aims to harness the potential of astrocytes, a type of brain cell often overlooked in AI development. By studying how astrocytes process information, researchers hope to create next-generation AI systems that learn faster and adapt more reliably.
AI models struggled to maintain focus on a task, degrading in accuracy as the word list length grew longer. Human performance, however, remained stable even with long lists, suggesting fundamental limitations in AI decision-making abilities.