The 46th COSPAR Scientific Assembly brought together experts from astronomy, the satellite industry, policy and law to examine one of the most important challenges arising from the rapid growth of activities in space: how to preserve dark and quiet skies while enabling satellite systems to continue delivering valuable services to society. The meeting was chaired by Marco Tavani. 

The exponential demand for energy-efficient and adaptive computing architectures drives the evolution of artificial intelligence (AI) and machine learning (ML). Neuromorphic computing, inspired by biological neural networks, overcomes the limitations of traditional von Neumann architectures, including high energy consumption and limited scalability. The introduction of two-dimensional (2D) materials, such as transition metal dichalcogenides, hexagonal boron nitride, black phosphorus, and tellurene, enables neuromorphic devices with unprecedented control over electronic and optoelectronic properties. These materials exhibit atomic-scale thickness, high carrier mobility, and tunable bandgaps, facilitating synaptic behaviours such as spike-timing-dependent plasticity and paired-pulse facilitation. This review describes the integration of 2D materials into neuromorphic systems, highlighting applications in wearable electronics, brain–machine interfaces, and quantum neuromorphic platforms. In wearable and edge computing, 2D-based devices enable localized, ultra-low-power data processing. In brain–machine interfaces, they enhance signal transduction and neural interfacing. Quantum effects in 2D materials further enable hybrid quantum–classical neuromorphic architectures for high-dimensional computational tasks. Despite significant advances, challenges in reproducibility, scalability, and stability remain. Addressing these limitations through innovations in synthesis and defect passivation is essential for practical application. This review underscores the transformative potential of 2D-material-based neuromorphic computing for energy-efficient AI.

Locating small lung tumors located deep within the lungs remains challenging, as they need to be navigated through a complex network of tiny airways. Existing AI systems trained on incomplete human annotations may also fail to identify peripheral airways that were omitted from the training labels. Researchers from Pusan National University developed a novel AI framework, ASTRA-Net, which helps recover previously overlooked peripheral airways—potentially enabling more accurate bronchoscopy maps for better lung cancer diagnosis.

Research data is the basis of scientific knowledge. However, valuable measurement results often remain unused after a project has ended or can hardly be related to data from other disciplines. The National Research Data Infrastructure (NFDI) sets standards and creates services so that researchers from different disciplines can easily find and reuse data. Knowledge is no longer lost, but can be incorporated into new research projects. This is why the German federal and state governments have decided to fund the ten consortia involved in the second NFDI funding round with roughly EUR 71 million until the end of 2028. The Karlsruhe Institute of Technology (KIT) participates in five of these consortia that manage data from individual disciplines. 

Large language models power today's AI applications, but generating responses quickly and efficiently remains a significant challenge. In a new study, researchers from Japan developed UniSpec, a training-free framework that speeds up LLM inference without retraining or altering model outputs. The team also introduced Multi-SpecBench, the first multilingual benchmark of its kind, demonstrating up to 2.6× faster inference across multiple LLM architectures, GPU platforms, and seven languages while producing outputs identical to standard autoregressive decoding. 

Researchers from the University of Osaka, Gunma University and Kyushu University created lipid mimic molecules that form “island”-like groups called lipid domains in artificial lipid membranes. The domains are thermally stable because the propellor-shaped moiety of the lipid mimic molecules interlock and the surrounding lipids fit into the spaces between. This new method of forming domains may allow artificial lipid membrane materials to be created for catalysis, capturing molecules, and drug delivery.