OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows

A new framework called Speculative Verdict (SV) has been introduced to enhance the reasoning capabilities of Vision-Language Models (VLMs) when dealing with complex, information-rich images. SV operates in two stages: the draft stage, where small VLMs generate diverse reasoning paths, and the verdict stage, where a stronger VLM synthesizes these paths to produce accurate answers efficiently.

Recent advancements in Vision-Language Models (VLMs) have led to the development of Training-free Dual Hyperbolic Adapters (T-DHA), a novel adaptation method that enhances cross-modal reasoning without requiring extensive training resources. This method utilizes hyperbolic space to better represent hierarchical relationships between semantic concepts, improving both representation and discrimination capabilities.

A new benchmark called Tri-Bench has been introduced to assess the reliability of Vision-Language Models (VLMs) in spatial reasoning tasks, particularly under conditions of camera tilt and object interference. The benchmark evaluates four recent VLMs using a fixed prompt and measures their accuracy against 3D ground truth, revealing an average accuracy of approximately 69%.

GeoShield has been introduced as a novel adversarial framework aimed at protecting geolocation privacy from Vision-Language Models (VLMs) like GPT-4o, which can infer users' locations from publicly shared images. This framework includes three modules designed to enhance the robustness of geoprivacy protection in real-world scenarios.

A new task called Cross-View Point Correspondence (CVPC) has been proposed to enhance spatial understanding in Vision-Language Models (VLMs). This task is supported by the introduction of CrossPoint-Bench, a benchmark designed to evaluate models based on human cognitive processes of perception, reasoning, and correspondence. The evaluation reveals that current state-of-the-art models, such as Gemini-2.5-Pro, significantly lag behind human performance, with a 54.65% accuracy gap.

The introduction of AutoNeural marks a significant advancement in the design of Vision-Language Models (VLMs) specifically optimized for Neural Processing Units (NPUs). This architecture addresses the inefficiencies of existing VLMs on edge AI hardware by utilizing a MobileNetV5-style backbone and integrating State-Space Model principles, enabling stable integer-only inference.

The introduction of MedGR$^2$, a novel framework for Generative Reward Learning in medical reasoning, addresses the critical shortage of high-quality, expert-annotated data that hampers the application of Vision-Language Models (VLMs) in medicine. This framework enables the automated creation of multi-modal medical data, enhancing the training process for both Supervised Fine-Tuning and Reinforcement Learning.

A novel approach has been proposed for Visual Question Answering (VQA) in autonomous driving, integrating Vision-Language Models (VLMs) with continual learning techniques. This framework addresses the challenge of catastrophic forgetting when models are exposed to new driving tasks, enhancing their ability to understand and reason about their surroundings.