The Virtual Research Environment is an analysis platform developed at CERN serving the needs of scientific communities involved in European Projects. Its scope is to facilitate the development of end-to-end physics workflows, providing researchers with access to an infrastructure and to the digital content necessary to produce and preserve a scientific result in compliance with FAIR principles. The platform's development is aimed at demonstrating how sciences spanning from High Energy Physics to Astrophysics could benefit from the usage of common technologies, initially born to satisfy CERN's exabyte-scale data management needs. The Virtual Research Environment's main components are (1) a federated distributed storage solution (the Data Lake), providing functionalities for data injection and replication through a Data Management framework (Rucio), (2) a computing cluster supplying the processing power to run full analyses with Reana, a re-analysis software, (3) a federated and reliable Authentication and Authorization layer and (4) an enhanced notebook interface with containerised environments to hide the infrastructure's complexity from the user. The deployment of the Virtual Research Environment is open-source and modular, in order to make it easily reproducible by partner institutions; it is publicly accessible and kept up to date by taking advantage of state of the art IT-infrastructure technologies.
Recent methods demonstrate that large-scale pretrained models, such as CLIP vision transformers, effectively detect AI-generated images (AIGIs) from unseen generative models when used as feature extractors. Many state-of-the-art methods for AI-generated image detection build upon the original CLIP-ViT to enhance this generalization. Since CLIP's release, numerous vision foundation models (VFMs) have emerged, incorporating architectural improvements and different training paradigms. Despite these advances, their potential for AIGI detection and AI image forensics remains largely unexplored. In this work, we present a comprehensive benchmark across multiple VFM families, covering diverse pretraining objectives, input resolutions, and model scales. We systematically evaluate their out-of-the-box performance for detecting fully-generated AI-images and AI-inpainted images, and discover that the best model outperforms the original CLIP by more than 12
Membrane protection against oxidative insults is achieved by the concerted action of glutathione peroxidase 4 (GPX4) and endogenous lipophilic antioxidants such as ubiquinone and vitamin E. More recently, ferroptosis suppressor protein 1 (FSP1) was identified as a critical ferroptosis inhibitor, acting via the regeneration of membrane-embedded antioxidants. Yet, regulators of FSP1 are largely uncharacterized, and their identification is essential for understanding the mechanisms buffering phospholipid peroxidation and ferroptosis. Here we report a focused CRISPR-Cas9 screen to uncover factors influencing FSP1 function, identifying riboflavin (vitamin B2) as a modulator of ferroptosis sensitivity. We demonstrate that riboflavin supports FSP1 stability and the recycling of lipid-soluble antioxidants, thereby mitigating phospholipid peroxidation. Furthermore, we show that the riboflavin antimetabolite roseoflavin markedly impairs FSP1 function and sensitizes cancer cells to ferroptosis. Our findings provide a rational strategy to modulate the FSP1-antioxidant recycling pathway and underscore the therapeutic potential of targeting riboflavin metabolism, with implications for understanding the interaction of nutrients, as well as their contributions to a cell's antioxidant capacity.
Hyperspectral imaging enables fine-grained recognition of materials by capturing rich spectral signatures, but learning robust classifiers is challenging due to high dimensionality, spectral redundancy, limited labeled data, and strong domain shifts. Beyond earth observation, labeled HSI data is often scarce and imbalanced, motivating compact models for generic hyperspectral classification across diverse acquisition regimes. We propose the lightweight Spectral-Spatial Fusion Transformer (SSFT), which factorizes representation learning into spectral and spatial pathways and integrates them via cross-attention to capture complementary wavelength-dependent and structural information. We evaluate our SSFT on the challenging HSI-Benchmark, a heterogeneous multi-dataset benchmark covering earth observation, fruit condition assessment, and fine-grained material recognition. SSFT achieves state-of-the-art overall performance, ranking first while using less than 2
Recent advances in AI-based image generation, particularly in diffusion-based methods, have produced powerful and easily-accessible tools that can be exploited for misinformation and fraud. Detecting diffusion-generated and -inpainted images, especially from rapidly evolving latent diffusion models (LDMs), remains highly challenging due to their ability to produce photorealistic forgeries. We present DualSight, a parallel architecture that disentangles semantic cues from generation artifacts through a dual-branch encoder and a progressive local-to-global feature decoder. The semantic branch leverages a CLIP-pretrained vision encoder fine-tuned via low-rank adaptation, ensuring strong generalization to unseen diffusion models. The artifact branch employs a hierarchical encoder trained on a content-agnostic image representation using static filters, capturing subtle generation artifacts. On the OpenSDI benchmark, DualSight surpasses the recent state-of-the-art by over 10 https://github.com/CeMOS-IS/dualsight .
