International multisite implementation of distributed cell-free protein biomanufacturing to advance health and research equity - Science | AAAS

image: Overview of memory-efficient training techniques for transformers The integration of deep learning—particularly Transformer-based models—has profoundly reshaped scientific research paradigms. In domains such as drug discovery, weather forecasting, and protein structure prediction, these models enable the extraction of complex patterns from large-scale data. However, as model resolution and parameter counts continue to grow, GPU memory consumption increases rapidly, creating a “memory wall” that confines large-scale training to costly, high-end hardware and limits broader adoption in scientific research.

image: Overview of memory-efficient training techniques for transformers The integration of deep learning—particularly Transformer-based models—has profoundly reshaped scientific research paradigms. In domains such as drug discovery, weather forecasting, and protein structure prediction, these models enable the extraction of complex patterns from large-scale data. However, as model resolution and parameter counts continue to grow, GPU memory consumption increases rapidly, creating a “memory wall” that confines large-scale training to costly, high-end hardware and limits broader adoption in scientific research.

We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply. AbstractExtreme precipitation events increasingly threaten historic buildings in river basins, yet existing studies lack a basin-scale perspective on nonlinear flood risk mechanisms. Taking the Minjiang River Basin as the study area, this research integrates hydrological simulation (SCS-CN), machine learning (XGBoost-SHAP), and spatial autocorrelation analysis to assess flood risk for 5482 historic buildings. Results reveal three key patterns: flood hazard decreases from downstream to upstream with belt-like aggregation along rivers; vulnerability shows a spatial mismatch, being higher upstream and lower downstream; integrated risk exhibits a southeast-high to northwest-low gradient with strong clustering in estuarine and midstream valleys. Key drivers, including impervious surfaces, vegetation cover, and topography, demonstrate significant nonlinear threshold effects and multi-factor coupling.

Discover the cosmos! Each day a different image or photograph of our fascinating universe is featured, along with a brief explanation written by a professional astronomer. X-ray: NASA/CXC/INAF-IRA/A. Botteon et al.; Infrared: UKIDSS-GPS; Radio: ASTRONLOFAR; Image processing: NASA/CXC/SAO/N. Wolk

A Johns Hopkins University scientist whose award-winning work suggests that a brief stomach bug could raise the risk of colon cancer has moved to China full time. Wan Fengyi, a professor at the Johns Hopkins Bloomberg School of Public Health, has joined Zhejiang University in Hangzhou, according to his new profile on the Chinese university’s website. Wan returned to China this month and now holds a chair professorship at Zhejiang University. He is based at a national laboratory at its Second...