Signals, in order.

Lorentz Local Canonicalization (LLoCa) provides exact Lorentz-equivariance for arbitrary neural networks with minimal computational overhead. It is applied to the LHC, predicting local reference frames for each particle and propagating any-order tensorial information between them, and has been tested on graph networks and transformers. The method demonstrates cutting‑edge performance on amplitude regression, end‑to‑end event generation, and jet tagging, and a large top tagging dataset was introduced to benchmark LLoCa versions of various benchmark architectures, highlighting the importance of…

Source: OpenAlex

TrackStudio is a modular, GUI-based pipeline that integrates open‑source tools for markerless motion tracking. It supports video recording preprocessing, synchronization, automatic 2D and 3D pose estimation, and visualization without programming. In tests with 76 participants across three environments, it achieved inter‑frame correlations above 0.98 and triangulation errors below 13.6 mm for hand tracking.

Source: OpenAlex

Novaya Politra, a Canadian student from a Bengali family, is studying landscape architecture at ETH Zurich. Despite a demanding workload, homesickness, and visits to devastated landscapes, she finds inspiration and fresh perspectives from her surroundings. Students are encouraged to use counselling services for support and can apply for the Excellence Scholarship & Opportunity Programme (ESOP) to pursue a Master’s degree at ETH Zurich.

Source: ETH Zurich News

The RobinCar Family consists of two R packages, RobinCar and RobinCar2, designed to perform covariate‑adjusted analyses for continuous, discrete, and time‑to‑event outcomes in clinical trials. The packages implement traditional methods such as ANCOVA and newer approaches like ANHECOVA, G‑computation, and super‑covariate adjustment, as well as time‑to‑event techniques including covariate‑adjusted log‑rank tests and hazard ratio estimation. The tools were demonstrated on data from the AIDS Clinical Trials Group Study 175, showing they are straightforward and user‑friendly.

Source: OpenAlex

The study assessed the accuracy and completeness of AI-generated guidance for inflammatory bowel disease (IBD) using GPT-4 and DeepSeek. Seven IBD specialists rated responses from 20 real-world questions and a standardized vignette, while 12 non-specialist physicians evaluated clinical usefulness and completed a two‑round case‑based test. Results showed both models produced generally accurate answers (median scores ≥4/5) and improved non‑specialist physicians’ test scores from 5.5–6.0 to 8.0, with no harmful recommendations identified.

Source: OpenAlex

Researchers from EPFL, ETH Zurich, and CSCS have partnered with Swiss company Proton to integrate the open‑source language model Apertus 1.5 into Proton’s AI assistant Lumo. The collaboration allows Apertus developers to receive voluntary, anonymised user feedback from millions of Lumo users, a first for an open model. The partnership also supports Switzerland’s goal of AI sovereignty by providing a competitive, privacy‑focused alternative to foreign chatbots.

Source: EPFL News

Researchers from EPFL, ETH Zurich, and the Swiss National Supercomputing Centre have partnered with Proton to make Apertus 1.5 available in Proton’s Lumo AI assistant. This integration allows Lumo users to choose Apertus as their language model and voluntarily share anonymised feedback, giving the Apertus team real‑world usage data for the first time. The partnership supports the development of open, sovereign AI in Switzerland and Europe.

Source: ETH Zurich News

Researchers from Empa and the University Children’s Hospital Zurich have launched a four‑year joint project to develop a non‑invasive diagnostic protocol for asthma in children as young as two years old. The protocol combines buccal DNA swabs with breath analysis via high‑resolution mass spectrometry, and employs machine‑learning models to identify asthma‑specific molecules and genetic variants. Successful completion could enable earlier treatment, reduce hospital admissions, and prevent unnecessary medication in non‑asthmatic children.

Source: Swiss Federal Government

Researchers at ETH Zurich and the University of Zurich have created a magnetically controllable, flexible fetoscope that can bend up to 173 degrees and provide a panoramic view for navigating blood vessels in twin-to-twin transfusion syndrome (TTTS). The system, which uses external electromagnets and a PlayStation controller for manual override, achieved a precision of 140 µm in experiments and successfully performed the key steps of a TTTS procedure in a pregnant ewe. The platform aims to make the complex laser surgery for TTTS safer, more precise, and more accessible to specialists.

Source: ETH Zurich News

The Center for Research on Learning, Understanding and Education (CLUE) was founded by ETH and EPFL to strengthen cooperation between research, schools and education authorities. It will connect existing learning sciences groups, facilitate collaborative projects, and involve teachers through a sabbatical program to ensure research questions are relevant to everyday school life. The center will also seek third‑party funding and host events where researchers and teachers can meet.

Source: ETH Zurich News