A team of scientists from Universidad Carlos III de Madrid (UC3M) and Harvard University have experimentally demonstrated that it is possible to reprogram the mechanical and structural behaviour of innovative artificial materials with magnetic properties, known as metamaterials, without the need to modify their composition. This technology opens the door to innovations in fields such as biomedicine and soft robotics, among others.

Researchers at Tongji University have unveiled a novel framework, the Global Confidence Degree-based Graph Neural Network (GCD-GNN), designed to significantly improve the detection of financial fraud in complex transaction networks. Published in Artificial Intelligence and Autonomous Systems (AIAS), this innovative approach addresses critical challenges in identifying fraudulent activities, such as camouflaged transactions and intricate relational patterns, by integrating global confidence metrics with advanced graph learning techniques.

As lithium-ion batteries (LIBs) continue to power electric vehicles and energy storage systems, their long-term health remains a critical challenge. A groundbreaking new method combines deep learning with physical modeling to deliver rapid, accurate degradation diagnosis at the electrode level. This innovative approach requires only 11 data points from a charging cycle, achieving reliable predictions in just 2.5 minutes. By simplifying the data needed and eliminating the need for specialized equipment, this technique offers a faster, more accessible solution to battery health monitoring—paving the way for safer, more efficient battery management in large-scale applications.

This study introduces a deep-learning system for rapid, automated detection and classification of tiny calcium deposits (microcalcifications) in mammograms to aid early breast cancer diagnosis. Leveraging a multi-center dataset of 4,810 biopsy-confirmed mammograms, our pipeline uses a Faster RCNN model with a feature-pyramid backbone to detect and classify microcalcifications—the pipeline requires no hand-tuned rules and provides both the overall cancer risk and highlighted lesion regions in seconds per image. On unseen test data, it achieved overall classification accuracy of 72% for discriminating between benign and malignant breasts and 78% sensitivity of malignant breast cancer prediction, marking a significant step toward AI-assisted, cost-effective breast-cancer screening that can run on standard radiology workstations.

To protect artists, University of Tennessee researchers have developed HarmonyCloak, a tool that prevents AI from reproducing an artist’s unique sound by embedding imperceptible noises in music data, making it unlearnable to AI models.

The Center for Coastal and Ocean Mapping at the University of New Hampshire has become the world leader in both how we map the seafloor and who conducts that exploration.