86 stories · Source dates are shown for news briefs.
NRKZ Signal · Signal illustrationResearchers developed an AI system that matches real-time X-ray images with preoperative 3D scans, making minimally invasive surgeries safer and more precise. The system, called xvr, adapts to each patient in just five minutes and outperforms existing AI methods by an order of magnitude. It could lead to faster and safer procedures.
NRKZ Signal · Signal illustrationGoogle researchers have developed a new framework called Retrieve-for-Train that bypasses the bottleneck of expensive inference-time reasoning. Instead, it uses reinforcement learning to train a lightweight diffusion model, allowing for the instant generation of a cohesive, expert-level set of AI search results. This approach enables modern search or recommendation applications to return a coherent set of results rather than a single best match, addressing the need for systems to break broad prompts into related sub-queries to cover potential user interests.
NRKZ Signal · Signal illustrationThe National Institute of Standards and Technology (NIST) has finalized guidelines to protect tokens and assertions used in digital systems. These tokens are crucial for maintaining security and preventing unauthorized access. The guidelines, titled 'Protecting Tokens and Assertions from Forgery, Theft, and Misuse,' are aimed at federal agencies and cloud service providers, but can also benefit any organization handling identity tokens and assertions.
NRKZ Signal · Signal illustrationMIT researchers have developed a new method that helps generative artificial intelligence models meet hard constraints in safety-critical situations. The technique, which allows more freedom during generation and enforces constraints on the final output, consistently outperformed existing methods in experiments across robotics, control of physical processes, and computer vision. The method ensures plausible outputs while satisfying nonnegotiable safety, physical, or task-specific requirements.
NRKZ Signal · Signal illustrationToolGrad is a data generation framework that first generates tool-use chains before annotating user queries. It enables LLMs to achieve better tool-use performance, outperforming baseline methods and proprietary LLMs on out-of-distribution datasets.
NRKZ Signal · Signal illustrationResearchers at MIT Media Lab have developed injectable nanoantennas that can create localized therapeutic electric fields to target and kill brain cancer cells without damaging healthy tissue. Laboratory and animal studies show significant reduction in tumor growth and extended survival, highlighting potential as a precise and safe brain cancer therapy. The technology, named HITMAN, is described in an open-access paper in Science Advances.
NRKZ Signal · Signal illustrationHow will global access to food, water, and energy evolve in coming decades?
NRKZ Signal · Signal illustrationA study evaluates how transfer learning from European cohorts improves genomic risk prediction in small underrepresented populations but degrades accuracy when target cohort sample sizes grow. The research suggests a potential solution for optimizing predictive performance in non-European populations by combining large-scale European GWAS results with targeted population-specific GWAS.
NRKZ Signal · Signal illustrationA new deep-learning framework called Methane Analysis and Plume Localization with EMIT (MPL) automates the detection, quantification, and source estimation of methane emissions globally. This technology, powered by satellite data, helps track localized point sources in waste, agriculture, and energy sectors, offering a critical
NRKZ Signal · Signal illustrationTimesFM-3 is a state-of-the-art time series foundation model that enables highly accurate multivariate time series forecasting in a single forward pass. It significantly outperforms other forecasting models across major benchmarks. Since its debut in 2024, TimesFM has been adopted for real-world time-series forecasting tasks across multiple domains. TimesFM-3 is pre-trained on a real-world and synthetic time-series corpus comprising more than 1 trillion time points and supports joint prediction of multiple coevolving time series without requiring task-specific fine-tuning.
NRKZ Signal · Signal illustrationA lightweight, self-supervised continuous glucose monitoring (CGM) foundation model called GlucoFM has been developed by Google Research. GlucoFM models slower glucose trends and short-term deviations separately, improving performance across diverse metabolic prediction tasks. Unlike existing models, GlucoFM separates CGM data into distinct streams, enabling better interpretation and analysis of glucose regulation.
NRKZ Signal · Signal illustrationAgentHands is an LLM-powered XR prototype that augments conversational agents with synchronized, expressive hand gestures to provide spatially grounded guidance, bridging the mental mapping gap and enhancing user engagement in physical tasks. It leverages the spatial understanding capabilities of XR to replicate natural co-speech gestures, making conversations about surroundings more intuitive and engaging.
NRKZ Signal · Signal illustrationAn AI system called Biomarker Discovery Framework prioritizes candidate biomarkers from wearable sensor data, addressing the bottleneck in turning physiological signals into reliable biomarkers. It combines hypothesis generation, statistical analysis, and literature-grounded reasoning to maintain strict statistical rigor and human oversight. Across three cohorts, it recovered known clinical signals, identified convergent biomarkers, and improved downstream prediction when combined with demographic features.
NRKZ Signal · Signal illustrationGoogle Research introduces a dynamic, mobility-informed framework that enhances AI models' understanding of places by incorporating real-world functional dynamics. The framework, called Mobility-Embedded POIs (ME-POIs), improves predictions on attributes like opening hours, price levels, and busyness. Using publicly available benchmark datasets, ME-POIs blend text descriptions with anonymized mobility patterns, resulting in a 81.9% relative gain in visit intent prediction, 75.1% improvement in price level classification, and 24.7% increase in busyness estimation accuracy across unseen places.
NRKZ Signal · Signal illustrationGoogle Research introduces PhotoScan, a deep learning approach that estimates body composition from smartphone photos, predicting insulin resistance with accuracy comparable to DXA scans. This complements wearable sensor data with objective measures of body composition, offering a complete picture of metabolic risk. While DXA scans are the gold standard, PhotoScan's accuracy surpasses BIA sensors, providing insights into A/G and V/S ratios beyond BIA's capabilities. The study validates PhotoScan's accuracy across clinical cohorts and demonstrates its feasibility for use in a clinical research setting.
NRKZ Signal · Signal illustrationWe then show that many factual errors in frontier LLMs are better understood as lost keys (recall failures) , not empty shelves (encoding failures) .
NRKZ Signal · Signal illustrationGoogle's AMIE AI system demonstrates expert-level performance in real-time video consultations, with a randomized controlled study showing its effectiveness in simulated consultations. The system integrates continuous visual and auditory cues with spoken clinical history, enhancing patient trust and communication. AMIE has been extended towards specialist-level evaluations in oncology, cardiology, and ophthalmology, and multimodal diagnostic reasoning over images and clinical documents. The system is currently being translated towards clinical practice through a framework for physician-centered oversight and ongoing randomized studies.
NRKZ Signal · Signal illustrationNIST researchers have developed a new analysis method to correct a pervasive error in data analysis of nanoparticles, which can provide misleading insights into their properties. The method, reported in ACS Nano, has implications for testing material properties and developing nanotechnology products. The correction reveals how nanoscale objects truly behave, including their color and other properties.
Follow our open news feed. No inbox clutter.
Just the latest dispatches, in your reader.
