News

A 31-billion-parameter open-weight multimodal model (Gemma 4 31B-IT) fine-tuned with supervised fine-tuning and retrieval-augmented generation using fixed-size chunking passed 8 of 14 U.S. Nuclear Regulatory Commission R…

Researchers introduce ConVBench, a benchmark for evaluating logical consistency in Large Vision-Language Models (LVLMs), pairing each image with two logically equivalent questions across six categories. They also present…

Researchers propose Risk-routed Implicit Boundary Refinement (RIBR), a compact segmentation framework that uses implicit neural representation as a risk-routed residual correction for robust ultrasound image segmentation…

Researchers propose SCALE, a self-supervised constraint-aware layout generation framework for local place-and-route design-rule violation fixing at sub-2nm nodes, boosting state-of-the-art agents' solve rates by +12–25% …

Researchers propose the Semantic Gaussian Allocation Transformer (SAGFormer), which uses Gaussian attributes and local geometric-semantic features to score candidates and select a fixed final Gaussian set for 3D semantic…

Researchers propose ISPCloak, an optimization-free adversarial attack framework that weaponizes Image Signal Processing (ISP) pipelines to fool deepfake detectors by imprinting hardware-intrinsic statistical signatures o…

A systematic review of 60 studies on diffusion models for medical image inpainting finds that denoising diffusion probabilistic models and latent diffusion models are the dominant architectures, primarily used for artifa…

Researchers propose TextSLIP, a medical vision-language pretraining framework that augments CLIP with intra-modal text contrastive learning to improve fine-grained semantic supervision for radiology report generation. Pr…

Researchers propose Visual Saliency Steering Distillation (VSSD) to improve multimodal chain-of-thought reasoning in small models by using attention maps from multimodal large language models to generate perturbed images…

Researchers propose EVL-MCoT, an enhanced vision-language multi-chain-of-thought approach for harmful meme detection, achieving promising results on the HatefulMemes and MultiOff datasets. The method addresses limitation…

A single DINOv3-pretrained ConvNeXt trained with a tuned recipe achieves 88.19 mAP on VeRi-Wild Small and 77.47 mAP on VeRi-Wild Large for vehicle re-identification, matching the strongest multi-branch baseline. Research…

A new pipeline using large language models to extract grammatical rules and lexicons from grammar books generates synthetic parallel corpora for fine-tuning machine translation models, improving over seed-data baselines …

A study testing four major LLM families (Claude, Grok, GPT, Gemini) on ethnonationalist pseudo-science found that Grok's Fast versions consistently assigned credibility scores of 70-75, two to five times higher than othe…

A new framework called MetaEvolve, developed by researchers and detailed in a paper on arXiv, teaches large language models to self-evolve through reinforcement learning, achieving a 10.01% absolute improvement on in-dis…

A new benchmark called MissionBench reveals that even the strongest multimodal large language models (MLLMs) succeed on fewer than 35% of aerial missions, compared to 84.4% human performance, according to a preprint on a…

Researchers from the University of Moratuwa have released grapheme-kit, an open-source Python library that computes lexical distance, similarity, and evaluation metrics on grapheme clusters rather than Unicode code point…

A new open-source framework from University College London (UCL) evaluates open-weight large language models on longitudinal data preparation tasks without sending data to the cloud. Benchmarking models up to 31-35B para…

A new study from arXiv presents a partial-identification approach to bounding the causal effect of a new ML model using prior randomized control trial data, leveraging assumptions about counterfactual correctness and sub…

A survey of 25 experts in requirements engineering, data science, development, and compliance finds that most organizations lack systematic processes to translate EU AI Act obligations into testable requirements and veri…

A new study from arXiv (2607.22218v1) finds that large language models (LLMs) generally rely on a narrower subset of human creativity evaluation standards, converging most with humans on novelty but diverging most on con…