Signals, in order.

The supplement details the parameters and spatial inputs of the SEALS v2.0 model, including a resolution pyramid, calibration loss illustration, full habitat‑quality validation, comparison with related downscaling models, and discussion of limitations and extensions. It describes calibration using the ESA 2000 land‑use map to match the 2015 ESA map, out‑of‑sample assessment on the 2020 map, and application of SSP‑RCP changes to the 2015 map for 2030, 2050, 2070, and 2100, with spatial‑suitability variables listed in Table S1 and adjacency effects modeled at two scales.

Source: OpenAlex

The study proposes a deep learning model to classify NATM excavation-support classes from tunnel face photographs. Using MobileNetV2 and MobileNetV3‑Small on 245 photos from the BT‑05 tunnel in Osmaniye, Turkey, the models achieved macro‑F1 scores of 0.893 and 0.847 respectively, with MobileNetV2 maintaining 0.844 accuracy on an independent test. Grad‑CAM analysis confirmed predictions were based on geologically relevant features, demonstrating a fast, objective decision‑support tool for on‑site geotechnical assessment.

Source: OpenAlex

Researchers at ETH Zurich and the Paul Scherrer Institute have produced a cold, high‑intensity muonium beam using superfluid helium, enabling precise tests of the universality of free fall for second‑generation particles. The experiment aims to measure how muonium, a neutral atom composed of a muon and an electron, falls under Earth’s gravity, potentially revealing deviations from Einstein’s equivalence principle. The team plans to use an interferometer to detect minute gravitational effects and also intends to apply the beam to laser spectroscopy for improved measurements of muon properties.

Source: ETH Zurich News

ETH Zurich’s Cleanroom Facility Basel and its sister facilities in Zurich provide state‑of‑the‑art environments for fabricating CMOS microelectronic chips, sensors, and semiconductor lasers. The rooms maintain strict temperature, humidity, and particle‑control standards, enabling researchers to produce structures as small as 500 nm and to test drug‑testing chips with human cells. The facilities are open to students, faculty, and industry partners, supporting a wide range of scientific projects.

Source: ETH Zurich News

The HPQ physics building at ETH Zurich’s Hönggerberg campus is under construction, designed to provide highly stable environments for sensitive experiments, including quantum computing research. The building will feature a six‑storey basement with vibration‑isolated labs, temperature control to 0.01 °C, and modular layouts to accommodate evolving research needs. Construction began in 2010, with the first basement level underway and the first floors expected to be completed by the end of 2024, and the building will house about 500 researchers and 22 professorships by 2030.

Source: ETH Zurich News

More than 12,000 students have returned to classes at EPFL in Lausanne, Switzerland, where artificial intelligence has become a fixture in university studies. A survey of student representatives shows that the vast majority use AI while recognizing its limitations, and teachers are exploring ways to incorporate it into courses. Learning experts emphasize the need for students to critically assess whether AI aids learning or does it for them.

Source: EPFL News

Researchers at EPFL identified four rubber-related chemicals in Swiss raw milk, with at least one compound present in seven of 17 samples. The study also found 14 such chemicals in dairy equipment components, indicating that food-contact materials may be a source of contamination. The findings, published in Food Chemistry X, highlight the need for further research into contamination pathways and potential health risks.

Source: EPFL News

AI models are increasingly solving long‑standing mathematical problems, exemplified by an OpenAI proof related to the Navier–Stokes equations, a Millennium Prize Problem. Fields Medalist Alessio Figalli and 24 other leading mathematicians warn that while AI can accelerate results, it does not replace the human processes of questioning, failure, and creative insight, and that the flood of AI‑generated publications may dilute genuine progress. They also highlight risks in education, where students may submit AI‑generated solutions without understanding the underlying reasoning, urging a shift t…

Source: ETH Zurich News

Women accounted for only 14.1% of founders of Swiss startups that raised at least one financing round in 2024, and female CEOs closed 10.5% of all funding rounds in 2025, while Swiss venture capital investment reached CHF 2.95 billion. The FE+MALE Think Tank report argues that systematic changes in how investors, accelerators and funding institutions evaluate, connect with and allocate capital can narrow this gap, citing case studies where process redesign increased female founder participation by up to 30% and doubled the proportion of female-led companies progressing to due diligence. Only …

Source: Startupticker

The I 3 LUNG study enrolled 2,396 patients in a real‑world, multimodal AI trial, integrating clinical and blood data, CT images, digital pathology, and genomics into machine‑learning early fusion (MLEF) and deep‑learning intermediate fusion (DLIF) models. Machine‑learning CB‑only models achieved AUCs up to 0.77 in the test set, outperforming traditional markers such as PD‑L1, ECOG PS, NLR, LDH, and LIPI, while external validation showed a performance drop (AUC 0.55–0.72). A clinical usability study demonstrated that both expert and non‑expert physicians improved their predictions when using t…

Source: OpenAlex