Distribution Matching Variational AutoEncoder

AraLingBench has been introduced as a fully human-annotated benchmark designed to evaluate the Arabic linguistic capabilities of large language models (LLMs). It encompasses five key categories: grammar, morphology, spelling, reading comprehension, and syntax, assessed through 150 expert-crafted multiple-choice questions. The evaluation of 35 Arabic and bilingual LLMs indicates that while these models show surface-level proficiency, they struggle with deeper grammatical and syntactic reasoning.

The introduction of Omniguard presents a novel approach to AI safety moderation, specifically targeting the detection of harmful prompts across various languages and modalities. This method enhances the accuracy of harmful prompt classification by 11.57% compared to existing baselines, addressing concerns about the misuse of large language models (LLMs) and their susceptibility to attacks that exploit language and modality mismatches.

GitHub has announced updates to its Copilot features, including new sharing capabilities in Copilot Spaces, an update for Visual Studio Copilot, and public preview access to OpenAI's GPT-5.1-Codex-Max model. These enhancements aim to improve user experience and collaboration among developers.

A novel segmentation framework called Neural Tissue Relation Modeling (NTRM) has been introduced to enhance histopathology image segmentation in non-melanoma skin cancer diagnostics. This framework integrates a tissue-level graph neural network with convolutional neural networks (CNNs) to better model spatial and functional relationships among tissue types, addressing challenges in regions with overlapping or morphologically similar tissues.

A new framework called Feature Auto-Encoder (FAE) has been introduced to adapt pre-trained visual representations for image generation, addressing challenges in aligning high-dimensional features with low-dimensional generative models. This approach aims to simplify the adaptation process, enhancing the efficiency and quality of generated images.

The Vision Action Transformer (VAT) has been introduced as an innovative architecture that enhances the capabilities of Vision Transformers (ViTs) by utilizing the full feature hierarchy, rather than just the final layer's features. This approach allows VAT to process specialized action tokens alongside visual features across all transformer layers, achieving a remarkable 98.15% success rate on LIBERO benchmarks in simulated manipulation tasks.

A new technical report on arXiv presents a concise implementation of diffusion models, emphasizing a code-execution perspective to bridge the gap between theoretical formulations and practical applications. The report includes approximately 300 lines of code that illustrate essential components such as forward diffusion, reverse sampling, and the training loop, aiming to enhance understanding among researchers.

GlimmerNet has been introduced as an ultra-lightweight convolutional network designed for UAV-based emergency monitoring, utilizing Grouped Dilated Depthwise Convolutions to achieve multi-scale feature extraction without increasing parameter costs. This innovative approach allows for effective global perception while maintaining computational efficiency, making it suitable for edge and mobile vision tasks.