A research paper by scientists at Beijing Friendship Hospital develop a systematic approach for automated ossicular-chain segmentation and measurement using ultra-high-resolution computed tomography (U-HRCT).

The research paper, published on Jul, 2, 2025 in the journal Cyborg and Bionic Systems.

For the first time worldwide, we have achieved remote, real-time control of fusion plasma using a digital twin running on a supercomputer located about 1,000 km away (round-trip network path ~2,000 km).

In magnetic confinement fusion power, sustaining and precisely controlling plasma at temperatures exceeding 100 million ℃ over long durations is essential. Yet “predicting-while-controlling” has been challenging due to model accuracy limits, computation speed, and unresolved physics. Our team has developed a system that applies data assimilation, continuously updating the predictive model with real-time measurements to improve accuracy and using accelerated parallel prediction to determine optimal unrehearsed control actions.

A research team from Kyoto University, the National Institute for Fusion Science (NIFS), the National Institutes for Quantum Science and Technology (QST), and the Institute of Statistical Mathematics (ISM), has connected the Large Helical Device (LHD) in Toki, Gifu, Japan to the new “Plasma Simulator” supercomputer in Rokkasho, Aomori, jointly procured by NIFS and QST, via the high-quality, high-bandwidth academic network SINET6. By exclusively using more than 20,000 Central Processing Unit (CPU) cores and minimizing communication latency, the team has realized real-time predictive control of LHD from a remote supercomputer. This approach — linking a large experimental facility and a large computing system over a ~2,000 km network loop — can serve as a foundation for real-time control beyond fusion.

Recently, Prof. Andrea Alù from the City University of New York and Dr. Guangwei Hu from Nanyang Technological University in Singapore summarized previous representative work in the field of terahertz topologies and reconfigurable metamaterial devices, discussed design and integration methods for existing reconfigurable terahertz topology platforms, and explored potential avenues for future research and development. The findings were published as the cover paper titled “Topological and Reconfigurable Terahertz Metadevices” in Research (Research, 2025 DOI: 10.34133/research.0882).

Mathematical reasoning is a fundamental aspect of intelligence, encompassing a spectrum from basic arithmetic to intricate problem-solving. Recent investigations into the mathematical abilities of large language models (LLMs) have yielded inconsistent and incomplete assessments. In response, we introduce MathEval, a comprehensive benchmark designed to methodically evaluate the mathematical problem-solving proficiency of LLMs in various contexts, adaptation strategies, and evaluation metrics. MathEval consolidates 22 distinct datasets, encompassing a broad spectrum of mathematical disciplines, languages (including English and Chinese), and problem categories (ranging from arithmetic and competitive mathematics to higher mathematics), with varying degrees of difficulty from elementary to advanced. To address the complexity of mathematical reasoning outputs and adapt to diverse models and prompts, we employ GPT-4 as an automated pipeline for answer extraction and comparison. Additionally, we trained a publicly available DeepSeek-LLM-7B-Base model using GPT-4 results, enabling precise answer validation without requiring GPT-4 access. To mitigate potential test data contamination and truly gauge progress, MathEval incorporates an annually refreshed set of problems from the latest Chinese National College Entrance Examination (Gaokao-2023, Gaokao-2024), thereby benchmarking genuine advancements in mathematical problem solving skills.

Using EUROSTAT data and double randomization, the co-led study improves the Benefit of the Doubt model through a novel Ensemble-DEA framework that mitigates the curse of dimensionality in SDG indicators. Published in Expert Systems with Applications, the method offers more reliable EU performance rankings and benchmarking tools for evaluating sustainability policies across member states.

Space manipulators play an important role in the on-orbit services and planetary surface operation whose reliability is a key issue. In the extreme environment of space, space manipulators are susceptible to a variety of unknown disturbances. Since it is difficult for the manipulator to be repaired immediately, once it fails, it will mean that it cannot complete the mission as expected, which may cause serious losses and dangers. How to have a resilient guarantee, i.e., the manipulator’s ability to resiliently recover and continue to complete tasks, in failure or disturbance is the core capability of its future development. The motion planning unit is used as the computing terminal of the manipulator’s joint motion trajectory. Compared with traditional motion planning, learning-based motion planning has gradually become a hot spot in current research. However, no matter what kind of research ideas, the single robotic manipulator is studied as an independent agent, making it unable to provide sufficient flexibility under conditions such as external force disturbance, observation noise, and mechanical failure.