Researchers introduce Mage-VL, an efficient codec-native streaming multimodal foundation model that reduces visual token consumption by over 75% while matching or outperforming flagship encoders. The model achieves up to…
Researchers propose CoSA, a training-free sparse attention method that couples a Kernel-Aware Proxy with an Ordered-Skipping Kernel to accelerate long-context inference. CoSA achieves a 4.93× attention speedup and reduce…
Researchers introduce PerceptionBench, a benchmark designed to evaluate atomic visual perception in multimodal large language models (MLLMs), finding that no model reaches 60% accuracy across ten atomic perceptual capabi…
A new study on arXiv (2607.24784v1) finds that large language models (LLMs) can assist specialised translators in finding equivalents from English to French but cannot replace specialised corpora. Researchers evaluated G…
Researchers have developed I-LW-DETR, the first fully integer-only lightweight detection transformer, enabling deployment on NPUs and microcontrollers by executing all operations including transformer nonlinearities in i…
Researchers propose PILA, a plug-and-play insertion module for LLM-native advertising that decouples ad insertion from content generation as a lightweight sidecar, enabling model-agnostic integration without compromising…
Researchers propose WHTMix, a Walsh-Hadamard token mixer that replaces global self-attention in stereo depth estimation transformers, reducing model compute by a factor of 2.46 and single-image inference latency by a fac…
A new study from arXiv proposes neuromorphic masked diffusion language models (N-MDLMs) that integrate block diffusion with spike-based neuromorphic computation to improve inference efficiency. The approach achieves subs…
A new arXiv preprint (2607.25202v1) presents a cross-lingual analysis of conversational entrainment in Mandarin-English, Hindi-English, and Spanish-English code-switched speech, finding that lexical entrainment generaliz…
Researchers developed PATHFinder Agent, an end-to-end conversational system that gathers patient health and social context through structured dialogue, curates individualized prenatal care plans aligned with the American…
Researchers propose UniMem, a self-routing framework for LLM agents that balances plasticity and stability in boundary-agnostic task streams by using learnable routing tokens to coordinate episodic and parametric memory …
Researchers introduce Diff-ID, a diffusion-based framework that enforces identity consistency in facial image generation while delivering photorealistic quality, using a custom 210K image dataset and integrating ArcFace …
Researchers propose OPERA, a multi-agent ensemble framework that treats expert weight assignment as an offline policy learning problem to address distribution shifts in biomedical image analysis. OPERA coordinates hetero…
MorphUNet, a diffusion-based face morphing framework introduced in a new arXiv preprint (2607.25092v1), achieves the best Morphing Attack Potential (MAP) of 0.919 on FEI and 0.886 on FRLL when at least three of six recog…
Researchers have released OpenPVMapper, a nationwide database of rooftop photovoltaic installations in mainland France containing 1,135,850 installations totaling approximately 15.01 GWp of installed capacity, built by a…
Researchers introduce IMPRINT, a zero-shot plug-and-play framework that enriches textual object queries with web-sourced images to improve grounding in queryable semantic maps for Object Goal Navigation (ObjectNav). The …
Researchers introduce OrganLens, a self-supervised method for organ-specific representation learning from CT scans, producing 11 organ-specific representations from a shared encoder without external segmentation masks. I…
Singapore-based startup Antomation has developed an AI-powered automated baggage loading system called Cargo Cobra that aims to speed up the job by 20% and eliminate the need for human baggage handlers to work inside air…
A 1.2B-parameter LLaMA-style causal model called TimeCapsule, trained exclusively on Victorian texts from 1800-1875, achieves a 45.4% perplexity reduction over a GPT-2 baseline on held-out Victorian prose, according to a…
A new study introducing CogArena, a 13-paradigm benchmark for evaluating cognitive ability structure in large language models, finds that across 55 open-weight models, paradigm correlations are positive and a common axis…