Researchers at the University of Virginia have developed a new way to identify scientific breakthroughs more accurately. In a study published in Science Advances, data science Ph.D. candidate Munjung Kim and collaborators introduce the Embedding Disruptiveness Measure (EDM), a citation-based metric designed to improve on the widely used Consolidation-Disruption (CD) Index.

The CD Index was created to distinguish truly disruptive research from highly cited but incremental work. However, the method can misclassify papers when multiple teams make similar discoveries at the same time. EDM uses machine-learning techniques inspired by neural language models to analyze how papers connect to both prior and subsequent research, making the measure more robust to citation patterns and better able to identify simultaneous breakthroughs.

The researchers found that EDM more reliably captures transformative scientific contributions, providing a stronger foundation for studies of innovation, scientific discovery, and research impact.

A new propulsion system combines the power and speed of conventional chemical thrusters with the precision and fuel-efficiency of electrical thrusters. The system could enable small satellites capable of both fast, powerful maneuvers and slower, precise adjustments.

Nearly 69 million people in the United States speak a language other than English at home, yet weather warnings have long been issued almost exclusively in English. A new study documents how the National Weather Service is using artificial intelligence to change that, developing a comprehensive translation program that delivers life-saving forecasts and alerts in Spanish, Chinese, Vietnamese, Samoan, French and other languages.

A new study from researchers at MUSC Hollings Cancer Center found that machine learning may help identify cancer patients at high risk of experiencing financial toxicity — the financial hardship and stress that often accompany cancer treatment. Using data from 793 patients, researchers tested six machine learning models and developed a web-based tool that can estimate a patient's likelihood of financial strain based on factors such as age, income, health status, treatment status and out-of-pocket costs. The best-performing model identified high-risk patients with 84% sensitivity and 75% specificity, offering a potential way for health care teams to intervene earlier with financial counseling and support services. The goal is to move financial toxicity screening from a reactive process to a proactive one, helping patients access resources before financial strain affects treatment decisions, quality of life or health outcomes.