Geometric Stability: The Missing Axis of Representations

Analysis of learned representations has a blind spot: it focuses on $similarity$, measuring how closely embeddings align with external references, but similarity reveals only what is represented, not whether that structure is robust. We introduce $geometric$ $stability$, a distinct dimension that quantifies how reliably representational geometry holds under perturbation, and present $Shesha$, a framework for measuring it. Across 2,463 configurations in seven domains, we show that stability and similarity are empirically uncorrelated ($ρ\approx 0.01$) and mechanistically distinct: similarity metrics collapse after removing the top principal components, while stability retains sensitivity to fine-grained manifold structure. This distinction yields actionable insights: for safety monitoring, stability acts as a functional geometric canary, detecting structural drift nearly 2$\times$ more sensitively than CKA while filtering out the non-functional noise that triggers false alarms in rigid distance metrics; for controllability, supervised stability predicts linear steerability ($ρ= 0.89$-$0.96$); for model selection, stability dissociates from transferability, revealing a geometric tax that transfer optimization incurs. Beyond machine learning, stability predicts CRISPR perturbation coherence and neural-behavioral coupling. By quantifying $how$ $reliably$ systems maintain structure, geometric stability provides a necessary complement to similarity for auditing representations across biological and computational systems.

MisSpans: Fine-Grained False Span Identification in Cross-Domain Fake News

Online misinformation is increasingly pervasive, yet most existing benchmarks and methods evaluate veracity at the level of whole claims or paragraphs using coarse binary labels, obscuring how true and false details often co-exist within single sentences. These simplifications also limit interpretability: global explanations cannot identify which specific segments are misleading or differentiate how a detail is false (e.g., distorted vs. fabricated). To address these gaps, we introduce MisSpans, the first multi-domain, human-annotated benchmark for span-level misinformation detection and analysis, consisting of paired real and fake news stories. MisSpans defines three complementary tasks: MisSpansIdentity for pinpointing false spans within sentences, MisSpansType for categorising false spans by misinformation type, and MisSpansExplanation for providing rationales grounded in identified spans. Together, these tasks enable fine-grained localisation, nuanced characterisation beyond true/false and actionable explanations. Expert annotators were guided by standardised guidelines and consistency checks, leading to high inter-annotator agreement. We evaluate 15 representative LLMs, including reasoning-enhanced and non-reasoning variants, under zero-shot and one-shot settings. Results reveal the challenging nature of fine-grained misinformation identification and analysis, and highlight the need for a deeper understanding of how performance may be influenced by multiple interacting factors, including model size and reasoning capabilities, along with domain-specific textual features. This project will be available at https://github.com/lzw108/MisSpans.

AI-Driven Evaluation of Surgical Skill via Action Recognition

The development of effective training and evaluation strategies is critical. Conventional methods for assessing surgical proficiency typically rely on expert supervision, either through onsite observation or retrospective analysis of recorded procedures. However, these approaches are inherently subjective, susceptible to inter-rater variability, and require substantial time and effort from expert surgeons. These demands are often impractical in low- and middle-income countries, thereby limiting the scalability and consistency of such methods across training programs. To address these limitations, we propose a novel AI-driven framework for the automated assessment of microanastomosis performance. The system integrates a video transformer architecture based on TimeSformer, improved with hierarchical temporal attention and weighted spatial attention mechanisms, to achieve accurate action recognition within surgical videos. Fine-grained motion features are then extracted using a YOLO-based object detection and tracking method, allowing for detailed analysis of instrument kinematics. Performance is evaluated along five aspects of microanastomosis skill, including overall action execution, motion quality during procedure-critical actions, and general instrument handling. Experimental validation using a dataset of 58 expert-annotated videos demonstrates the effectiveness of the system, achieving 87.7% frame-level accuracy in action segmentation that increased to 93.62% with post-processing, and an average classification accuracy of 76% in replicating expert assessments across all skill aspects. These findings highlight the system's potential to provide objective, consistent, and interpretable feedback, thereby enabling more standardized, data-driven training and evaluation in surgical education.

Focal-RegionFace: Generating Fine-Grained Multi-attribute Descriptions for Arbitrarily Selected Face Focal Regions

In this paper, we introduce an underexplored problem in facial analysis: generating and recognizing multi-attribute natural language descriptions, containing facial action units (AUs), emotional states, and age estimation, for arbitrarily selected face regions (termed FaceFocalDesc). We argue that the system's ability to focus on individual facial areas leads to better understanding and control. To achieve this capability, we construct a new multi-attribute description dataset for arbitrarily selected face regions, providing rich region-level annotations and natural language descriptions. Further, we propose a fine-tuned vision-language model based on Qwen2.5-VL, called Focal-RegionFace for facial state analysis, which incrementally refines its focus on localized facial features through multiple progressively fine-tuning stages, resulting in interpretable age estimation, FAU and emotion detection. Experimental results show that Focal-RegionFace achieves the best performance on the new benchmark in terms of traditional and widely used metrics, as well as new proposed metrics. This fully verifies its effectiveness and versatility in fine-grained multi-attribute face region-focal analysis scenarios.

SliceLens: Fine-Grained and Grounded Error Slice Discovery for Multi-Instance Vision Tasks

Systematic failures of computer vision models on subsets with coherent visual patterns, known as error slices, pose a critical challenge for robust model evaluation. Existing slice discovery methods are primarily developed for image classification, limiting their applicability to multi-instance tasks such as detection, segmentation, and pose estimation. In real-world scenarios, error slices often arise from corner cases involving complex visual relationships, where existing instance-level approaches lacking fine-grained reasoning struggle to yield meaningful insights. Moreover, current benchmarks are typically tailored to specific algorithms or biased toward image classification, with artificial ground truth that fails to reflect real model failures. To address these limitations, we propose SliceLens, a hypothesis-driven framework that leverages LLMs and VLMs to generate and verify diverse failure hypotheses through grounded visual reasoning, enabling reliable identification of fine-grained and interpretable error slices. We further introduce FeSD (Fine-grained Slice Discovery), the first benchmark specifically designed for evaluating fine-grained error slice discovery across instance-level vision tasks, featuring expert-annotated and carefully refined ground-truth slices with precise grounding to local error regions. Extensive experiments on both existing benchmarks and FeSD demonstrate that SliceLens achieves state-of-the-art performance, improving Precision@10 by 0.42 (0.73 vs. 0.31) on FeSD, and identifies interpretable slices that facilitate actionable model improvements, as validated through model repair experiments.

DarkPatterns-LLM: A Multi-Layer Benchmark for Detecting Manipulative and Harmful AI Behavior

The proliferation of Large Language Models (LLMs) has intensified concerns about manipulative or deceptive behaviors that can undermine user autonomy, trust, and well-being. Existing safety benchmarks predominantly rely on coarse binary labels and fail to capture the nuanced psychological and social mechanisms constituting manipulation. We introduce \textbf{DarkPatterns-LLM}, a comprehensive benchmark dataset and diagnostic framework for fine-grained assessment of manipulative content in LLM outputs across seven harm categories: Legal/Power, Psychological, Emotional, Physical, Autonomy, Economic, and Societal Harm. Our framework implements a four-layer analytical pipeline comprising Multi-Granular Detection (MGD), Multi-Scale Intent Analysis (MSIAN), Threat Harmonization Protocol (THP), and Deep Contextual Risk Alignment (DCRA). The dataset contains 401 meticulously curated examples with instruction-response pairs and expert annotations. Through evaluation of state-of-the-art models including GPT-4, Claude 3.5, and LLaMA-3-70B, we observe significant performance disparities (65.2\%--89.7\%) and consistent weaknesses in detecting autonomy-undermining patterns. DarkPatterns-LLM establishes the first standardized, multi-dimensional benchmark for manipulation detection in LLMs, offering actionable diagnostics toward more trustworthy AI systems.

Language-Guided Grasp Detection with Coarse-to-Fine Learning for Robotic Manipulation

Grasping is one of the most fundamental challenging capabilities in robotic manipulation, especially in unstructured, cluttered, and semantically diverse environments. Recent researches have increasingly explored language-guided manipulation, where robots not only perceive the scene but also interpret task-relevant natural language instructions. However, existing language-conditioned grasping methods typically rely on shallow fusion strategies, leading to limited semantic grounding and weak alignment between linguistic intent and visual grasp reasoning.In this work, we propose Language-Guided Grasp Detection (LGGD) with a coarse-to-fine learning paradigm for robotic manipulation. LGGD leverages CLIP-based visual and textual embeddings within a hierarchical cross-modal fusion pipeline, progressively injecting linguistic cues into the visual feature reconstruction process. This design enables fine-grained visual-semantic alignment and improves the feasibility of the predicted grasps with respect to task instructions. In addition, we introduce a language-conditioned dynamic convolution head (LDCH) that mixes multiple convolution experts based on sentence-level features, enabling instruction-adaptive coarse mask and grasp predictions. A final refinement module further enhances grasp consistency and robustness in complex scenes.Experiments on the OCID-VLG and Grasp-Anything++ datasets show that LGGD surpasses existing language-guided grasping methods, exhibiting strong generalization to unseen objects and diverse language queries. Moreover, deployment on a real robotic platform demonstrates the practical effectiveness of our approach in executing accurate, instruction-conditioned grasp actions. The code will be released publicly upon acceptance.

Learning Semantic Atomic Skills for Multi-Task Robotic Manipulation

While imitation learning has shown impressive results in single-task robot manipulation, scaling it to multi-task settings remains a fundamental challenge due to issues such as suboptimal demonstrations, trajectory noise, and behavioral multi-modality. Existing skill-based methods attempt to address this by decomposing actions into reusable abstractions, but they often rely on fixed-length segmentation or environmental priors that limit semantic consistency and cross-task generalization. In this work, we propose AtomSkill, a novel multi-task imitation learning framework that learns and leverages a structured Atomic Skill Space for composable robot manipulation. Our approach is built on two key technical contributions. First, we construct a Semantically Grounded Atomic Skill Library by partitioning demonstrations into variable-length skills using gripper-state keyframe detection and vision-language model annotation. A contrastive learning objective ensures the resulting skill embeddings are both semantically consistent and temporally coherent. Second, we propose an Action Generation module with Keypose Imagination, which jointly predicts a skill's long-horizon terminal keypose and its immediate action sequence. This enables the policy to reason about overarching motion goals and fine-grained control simultaneously, facilitating robust skill chaining. Extensive experiments in simulated and real-world environments show that AtomSkill consistently outperforms state-of-the-art methods across diverse manipulation tasks.

Video Reality Test: Can AI-Generated ASMR Videos fool VLMs and Humans?

Recent advances in video generation have produced vivid content that are often indistinguishable from real videos, making AI-generated video detection an emerging societal challenge. Prior AIGC detection benchmarks mostly evaluate video without audio, target broad narrative domains, and focus on classification solely. Yet it remains unclear whether state-of-the-art video generation models can produce immersive, audio-paired videos that reliably deceive humans and VLMs. To this end, we introduce Video Reality Test, an ASMR-sourced video benchmark suite for testing perceptual realism under tight audio-visual coupling, featuring the following dimensions: (i) Immersive ASMR video-audio sources. Built on carefully curated real ASMR videos, the benchmark targets fine-grained action-object interactions with diversity across objects, actions, and backgrounds. (ii) Peer-Review evaluation. An adversarial creator-reviewer protocol where video generation models act as creators aiming to fool reviewers, while VLMs serve as reviewers seeking to identify fakeness. Our experimental findings show: The best creator Veo3.1-Fast even fools most VLMs: the strongest reviewer (Gemini 2.5-Pro) achieves only 56% accuracy (random 50%), far below that of human experts (81.25%). Adding audio improves real-fake discrimination, yet superficial cues such as watermarks can still significantly mislead models. These findings delineate the current boundary of video generation realism and expose limitations of VLMs in perceptual fidelity and audio-visual consistency. Our code is available at https://github.com/video-reality-test/video-reality-test.

VLA-R: Vision-Language Action Retrieval toward Open-World End-to-End Autonomous Driving

Exploring open-world situations in an end-to-end manner is a promising yet challenging task due to the need for strong generalization capabilities. In particular, end-to-end autonomous driving in unstructured outdoor environments often encounters conditions that were unfamiliar during training. In this work, we present Vision-Language Action Retrieval (VLA-R), an open-world end-to-end autonomous driving (OW-E2EAD) framework that integrates open-world perception with a novel vision-action retrieval paradigm. We leverage a frozen vision-language model for open-world detection and segmentation to obtain multi-scale, prompt-guided, and interpretable perception features without domain-specific tuning. A Q-Former bottleneck aggregates fine-grained visual representations with language-aligned visual features, bridging perception and action domains. To learn transferable driving behaviors, we introduce a vision-action contrastive learning scheme that aligns vision-language and action embeddings for effective open-world reasoning and action retrieval. Our experiments on a real-world robotic platform demonstrate strong generalization and exploratory performance in unstructured, unseen environments, even with limited data. Demo videos are provided in the supplementary material.