Searcharxiv⌕ Search

arXiv subjects

Chao Shen

Publications and source records attributed to Chao Shen.

At least 37 records · Page 2Linked to original sources
arXiv2026

Where Agent Frameworks Fall Short: Examining Functional Challenges and Usability Concerns

Large language model (LLM) agents are increasingly built on agent frameworks that provide reusable abstractions for workflow orchestration, state management, tool integration, and execution control. However, the quality of this infrastructure layer remains insufficiently understood, particularly its functionality challenges and usability concerns, as existing studies have mainly examined traditional deep learning (DL) frameworks or model-level agent failures. Therefore, we conduct an empirical study of 5,669 bug reports and 809 feature requests from five mainstream agent frameworks: AutoGen, CrewAI, LangChain, LangGraph, and MetaGPT. We construct a four-dimensional taxonomy covering 22 root causes, seven symptoms, 11 motivations, and six requirements, and map them to the five-stage agent lifecycle. Across the four RQs, results show an execution centered quality pattern shaped by semantic interface boundaries. Reported bugs mainly manifest as Incorrect Functionality (76.00%) and involve more API, configuration, parsing, and serialization related causes than DL framework bugs, while their associations remain sparse and stage specific. Feature requests mainly target Feature Enhancement (49.07%) and reveal structured needs for Orchestration Expressiveness, Development Delivery, Model Adaptation, and Tool Ecosystem. These findings call for quality assurance beyond crash based and tensor level testing, with emphasis on API sequences, structured LLM outputs, serialization boundaries, execution traces, and execution centered maintenance, offering empirical guidance for reliable and usable agent framework infrastructure.

cs.SE↗
arXiv2026

MMAligner: Safeguarding Multimodal Large Language Models through Representation Calibration

Multimodal large language models (MLLMs) often refuse unsafe text prompts yet generate harmful responses to semantically equivalent multimodal inputs. Existing defenses either rely on external guardrails, which add inference overhead without repairing intrinsic flaws, or safety fine-tuning, which treats alignment as black-box optimization and may sacrifice utility or require large multimodal datasets. To identify the cause of this safety disparity, we analyze MLLM representations geometrically. We find that safety mechanisms learned from text persist across modalities: a shared safety subspace and refusal boundary remain effective, and representations inside this boundary consistently trigger refusals. However, unsafe multimodal inputs undergo a representation shift that places most of them outside the boundary, allowing them to bypass the model's intrinsic safety mechanism. This indicates that multimodal safety degradation stems from representation misalignment rather than the absence of safety capability. Based on this finding, we propose MMAligner, a safeguarding method that calibrates unsafe multimodal representations into the pre-existing refusal region. MMAligner applies a hard lower bound to ensure refusal, a soft upper bound to avoid excessive modification, and a preservation objective for benign inputs. Experiments across multiple open-source MLLMs show that MMAligner raises the average refusal rate on unsafe multimodal inputs to 99% with less than 2% utility degradation and minimal training data, substantially improving the safety-utility trade-off over existing baselines. (*Due to the notification from arXiv, "The Abstract field cannot be longer than 1,920 characters", the Abstract that appeared is shortened.)

cs.CR↗
arXiv2026

Evaluation-Verification Reward for Consistent Multi-Reference Image Editing

While recent image editing models have made rapid progress, multi-reference editing remains challenging, particularly in maintaining visual consistency across references and ensuring overall visual harmony. Reinforcement learning has proven highly effective for text-to-image generation and single-image editing, but its extension to multi-reference editing is hindered by the absence of suitable reward models that capture multi-image relational constraints. Moreover, naively using multimodal large language models(MLLMs) as zero-shot evaluators faces a key tension between hallucination-prone long-form reasoning and the limited deductive power of short-form judgments. We address these issues with a Multi-dimensional Evaluation-Verification Reward(EVR). EVR decomposes evaluation into distinct visual criteria; for each criterion, an MLLM Evaluator generates multiple candidate hypotheses, and a Verifier grounds each claim in concrete visual evidence to accept or reject it, producing reliable and fine-grained reward signals. Together with a scalable data pipeline, our method enables RL fine-tuning of off-the-shelf editors without architectural changes. Extensive experiments show substantial gains over the base Qwen-Image-Edit, improving consistency and harmony to match or surpass NanoBanana.

cs.CV↗
arXiv2026

HIERA: Hierarchical Multi-Agent Relevance Assessment for Content Discovery Systems

Content discovery systems depend on relevance judgment for search quality evaluation, but human annotation faces inter-annotator disagreement and scaling costs. While Large Language Models show promise as automated assessors, current approaches rely on flat aggregation strategies: single-step prompting, voting ensembles, or uncoordinated multi-agent pipelines that aggregate independent outputs without integration. We propose HIERA, a hierarchical multi-agent relevance assessment framework with four specialized agents: a Relevance Judge, Query Analyzer, Item Analyzer, and Relation Analyzer. The Judge determines when specialist analysis is needed; the Relation Analyzer then coordinates query and item analyses with external knowledge to establish relevance relationships for final judgment. Ablation studies show that the same agents and external knowledge without hierarchical coordination degrade performance, confirming that the coordination structure itself accounts for the improvement. Evaluation across five datasets (EVS, MSRD, ESCI, WANDS, Home Depot) shows improvements over 11 baselines: 10.2\% on Home Depot, 4.8\% on ESCI, and up to 38\% on EVS ($p < 0.05$). Hierarchical coordination yields 12.7\% improvement over uncoordinated collaboration using identical agents.

cs.MA↗
arXiv2026

Alignment Is Local: A Paired Diagnostic for GUI Agents under User-Side Persuasion

Trustworthy deployment of GUI agents in ubiquitous computing settings requires alignment that survives dynamic interaction and precise threat conditions, not just single-turn refusal of explicit harmful requests. We argue that prompt-level alignment, the dominant lightweight defense in current mobile agents, is a local phenomenon: it works reliably only in the narrow evaluation slice where it is typically measured, namely single-turn, explicitly-verbalized intent, and degrades systematically along two axes that any real user can traverse. Using a paired diagnostic on three frontier GUI agents, screen-grounded, user-side persuasion, with no environment injection, we show that a one-line guardrail achieves large single-shot ASR reductions, up to roughly 40 points, at near-zero over-refusal cost. Nevertheless, moving from independent probes to four-turn escalation chains raises guarded ASR by approximately 20 points on every model. Relative to the neutral baselines, this increase reflects substantial guardrail erosion for Qwen but a largely defense-orthogonal dynamic risk for Claude and GPT. The sign of the salience gap flips under the guardrail: concealed requests are not systematically more successful than explicit ones without a guardrail, but are more successful with one, indicating that the defense engages primarily when intent is named. Static single-turn ASR therefore overstates deployed robustness by a systematic and predictable margin.

cs.CR↗
arXiv2026

Open-AoE: An Open Egocentric Manipulation Dataset and Toolchain for Embodied Learning

Egocentric videos of human manipulation provide scalable supervision for embodied intelligence, yet existing resources rarely combine low-cost continuous capture, manipulation-level structured annotations, and reusable tools for robot learning. We present Open-AoE, an open, community-oriented egocentric manipulation dataset and toolchain spanning the full pipeline from smartphone capture to model training. Its first release contains approximately 2,000 hours of manipulation video collected in natural environments by 500+ contributors using 400+ smartphones. The dataset provides text annotations, MANO-based hand poses, camera trajectories, and temporally localized atomic actions. Open-AoE further includes a data processing pipeline that transforms raw recordings into structured samples through temporal action segmentation, semantic annotation, hand reconstruction, and camera trajectory reconstruction. Meanwhile, we provide a separate downstream toolchain supports visualization, cross-embodiment retargeting, model-specific data conversion, and training recipes for VLA policies, WAMs, and World Models. By integrating scalable capture, structured processing, and downstream adaptation, Open-AoE reduces the barriers to both data contribution and reuse, providing practical open infrastructure for embodied model training, human-to-robot transfer, and world modeling.

cs.RO↗
arXiv2026

On Success and Simplicity: A Second Look at Transferable Vision-Language Attack Pipeline

Vision-Language Pre-training Models (VLPMs) are known to be vulnerable to adversarial attacks. Recent transferable attacks on VLPMs have followed a common pipeline with complicated loss functions or multi-stage text/image attacks. However, in this paper, we demonstrate that such a sophisticated attack pipeline can be simpler yet more successful. Specifically, we identify three previously overlooked issues caused by inappropriate cross-modal interactions and excessive operations. To address them, we propose the Simple Vision-Language Attack (SimVLA) pipeline, which observably improves transferability and efficiency. Experiments on four datasets and three downstream tasks validate the superiority of our pipeline. For instance, on Flickr30k text-image retrieval dataset, our SimVLA outperforms the SOTA baseline in R@1 transferability by 8.01\%-14.71\%, while consuming only about 35.73\% of the time and 46.26\% of the max VRAM. Overall, the superiority of our SimVLA highlights the importance of leveraging domain knowledge (e.g., our proposed cross-modal word identification), while blindly pursuing intricate operations (e.g, complex loss functions and redundant multi-stage designs) may even be harmful. We hope our SimVLA can serve as a simple yet effective backbone for future extensions. Code is available at https://github.com/RYC-98/SimVLA.

cs.CV↗
arXiv2026

A Novel Method for Differential-Algebraic Dynamic Model Discovery in Power Systems: An LLM-Based Multi-Agent Collaborative Framework

With large-scale integration of emerging power electronic devices represented by grid-forming inverters, power system dynamics increasingly exhibit strong nonlinearity, multi-timescale coupling, and black-box control logic. These features hinder conventional parameter identification requiring known model structures and structure identification based on predefined function libraries, making complete differential-algebraic dynamic model recovery difficult under weak prior information. To address this challenge, this paper proposes an LLM-based multi-agent collaborative framework for differential-algebraic dynamic model discovery in power systems. It integrates heterogeneous exploratory agents, individual candidate model memories, parameter fitting and evaluation, and a coordinator agent. Under unified measurement-data constraints, agents generate candidate equation structures in parallel, while candidates are optimized, evaluated, retained, and summarized to provide closed-loop search guidance. The task is decomposed into differential equation structure discovery and algebraic closure discovery, enabling joint recovery of state dynamics, algebraic constraints, and key intermediate variables with incomplete prior information. Case studies on synchronous generators and grid-forming inverters show that the proposed method outperforms single-agent LLM-based discovery and conventional symbolic regression in reconstruction accuracy, generalization, search efficiency, and noise robustness. In the generator case, OOD MAPE reaches 0.19\%; in the inverter case, discovery time is reduced by 25.7\% compared with the single-agent LLM baseline.

eess.SY↗
arXiv2026

Rethinking Technology Stack Selection with AI Coding Proficiency

Large language models (LLMs) are now an integral part of software development workflows and are reshaping the whole process. However, existing technology selection methods mainly focus on the inherent attributes of technologies, overlooking whether the LLM can effectively leverage the chosen technology. Therefore, teams using LLM assistants risk choosing technologies that cannot be used effectively by LLMs, yielding high debugging effort and mounting technical debt. We foresee a practical question in the LLM era, is a technology ready for AI-assisted development? In this paper, we first propose the concept, AI coding proficiency, the degree to which LLMs can utilize a given technology to generate high-quality code snippets. We then conduct the first large-scale empirical study examining AI coding proficiency across 170 third-party libraries and six LLMs. Our findings reveal that libraries with similar functionalities can exhibit up to 84% differences in the quality score of LLM-generated code. These gaps can be translated into real engineering costs and steer developer choices toward a narrow set of technologies, threatening technological diversity in the ecosystem. We call on the community to integrate AI coding proficiency into technology selection frameworks and develop mitigation strategies, preserving competitive balance in AI-driven development.

cs.SE↗
arXiv2026

Band bases as common triangular bases in cluster algebras from surfaces

We consider the skein algebra of an unpunctured marked surface. Thurston previously constructed its band basis topologically. We show that this band basis coincides with the common triangular basis, which is a Kazhdan-Lusztig type basis for quantum cluster algebras analogous to the dual canonical basis of quantum groups. Our result confirms a conjecture of Thurston. It also provides new cases for the existence of the common triangular basis. In addition, we discover a phenomenon where certain unknots are arranged in configurations resembling beads on a necklace.

math.QA↗
arXiv2026

When Safe Concepts Become Unsafe: Multi-Concept Compositional Vulnerabilities in Text-to-Image Models

Text-to-image (T2I) models are increasingly optimized for following user instructions faithfully. However, we find that this capability introduces a safety vulnerability we call Multi-Concept Compositional Unsafety (MCCU). MCCU occurs when multiple individually safe concepts, if combined in a single generation request, lead to harmful or sensitive visual outputs. Unlike prior jailbreak settings, MCCU does not rely on adversarial prompts, model access, or explicitly disallowed content. Instead, the risk emerges from how the model composes multiple safe visual concepts into a single scene. To systematically measure this threat, we build TwoHamsters, a large-scale evaluation framework consisting of 20k prompts, 51 curated concept pairs, and six risk categories. We evaluate 13 T2I models under a black-box setting. Our results show a clear conflict between instruction-following and safety: models that follow prompts more faithfully tend to produce more MCCU failures. For example, FLUX.1 achieves a 99.35% Unsafe Alignment Rate while only reaching a 1.57% MCCU Defense Rate. We further evaluate three representative defenses, including safety filtering, MCCU-specific detector fine-tuning, and concept erasure, all of which fail against unseen concept combinations. Our findings suggest that compositional reasoning in T2I models creates an attack surface that is not captured by existing safety mechanisms. We anticipate the release of TwoHamsters will catalyze community development of advanced generative defense mechanisms.

cs.CR↗
arXiv2026

The iterated geometric Green's formula

Fang, Lan, and Xiao established the geometric Green's formula as a categorical isomorphism for arbitrary semisimple complexes. In this short note, we generalize their work to multi-step compositions. Specifically, we establish the iterated geometric Green's formulas for the composition of an $(n-1)$-fold restriction and an induction, as well as its dual.

math.RT↗
arXiv2026

VOID: Defeating Unauthorized Mimicry in Latent Diffusion Models

While Latent Diffusion Models (LDMs) have revolutionized visual synthesis, they are increasingly exploited for unauthorized mimicry of individuals. Existing defenses inject deceptive perturbations to steer the generated images toward irrelevant targets. However, this approach hinges on an ungrounded assumption: subtle perturbations can maintain their deceptive efficacy throughout an LDM's extensive generation process. In reality, the model's innate restoration mechanism will remove such perturbations and cause individual identities to re-emerge in the images generated. We propose VOID, a defense framework that overcomes this conundrum by manipulating an LDM's intrinsic stochasticity. VOID perturbs the diffusion pipeline in two novel ways: 1) amplifying the latent encoding errors to shatter an image's semantic structure, and 2) counteracting the target guidance signals to suppress the model's restoration capabilities. This results in a semantic corruption that thwarts any unauthorized mimicry. Notably, the security gain does not come at the price of visual utility, as VOID simultaneously manages to confine perturbations to human-imperceptible regions of protected images. Our comprehensive evaluation of 24 state-of-the-art defenses against 10 mimicry attacks on 5 datasets demonstrates VOID's unprecedented protection power: it increases the average Frechet Inception Distance (FID) from 113 to 365, a 223% improvement over the strongest defense to date.

cs.CV↗
arXiv2026

HDSL: A Hierarchical Domain-Specific Language for Structured 3D Indoor Scene Generation and Localized Editing with LLM Agents

Text-driven indoor scene generation and editing require an intermediate representation that language models can both produce and revise. Existing LLM-based systems often rely on scene graphs or global constraint lists, which are compact but underspecify local geometry and make instruction-based edits difficult to localize. We frame this problem as structured program generation and local program repair, and propose Hierarchical Descriptive Scene Language (HDSL), an XML/CSS-style domain-specific language for structured 3D indoor scenes. HDSL represents rooms, regions, objects, and support surfaces as a tree with local coordinates, making complex scenes easier to plan recursively and easier to retrieve for editing. Our pipeline uses LLM agents to generate HDSL subtrees with bounded verification, grounds non-virtual nodes through multimodal asset retrieval, and applies force-directed layout optimization to repair boundary and collision errors. For editing, Hierarchical Retrieval-Augmented Generation retrieves the relevant subtree, asks the LLM to rewrite only that local context, and merges the result back through a deterministic three-way merge. In our reproduced benchmark, HDSL improves average object coverage, text-scene alignment, and generation time over full text-to-scene baselines while remaining competitive with recent layout-only reproductions on geometry metrics; for editing, HRAG reduces token use by $5.22\times$ and runtime by $6.19\times$, produces valid DSL for all eight paired edits, and better preserves unrelated scene objects.

cs.CV↗
arXiv2026

Can Reasoning Path still be Effective as Input? Bridging Post-Reasoning to Chain-of-Thought Compression

Recent developments have enabled advanced reasoning in Large Language Models (LLMs) via long Chain-of-Thought (CoT), trading efficiency during inference for performance. Existing works focus on compressing generated CoT in reasoning, which impairs the necessary information for deriving the correct answer. In this work, we propose post-reasoning, a reasoning paradigm that takes CoT as a part of context to simplify the reasoning task for LLMs. We find that post-reasoning significantly reduces the generation length of LLMs, but its effectiveness hinges on the efficiency and the reliability of the contextual CoT generation. Therefore, we propose Upfront CoT (UCoT), an efficient post-reasoning framework for CoT compression. UCoT trains a lightweight model (compressor) to provide contextual CoT in form of soft tokens and trains the LLM (executor) to leverage this contextual CoT for producing the final answer. Extensive experiments show that UCoT maintains the powerful reasoning ability of executor while significantly reducing the length of CoT. It is worth mentioning that when applying UCoT to the Qwen2.5-7B-Instruct model, the usage of tokens on GSM8K dataset is reduced by 50%, while the performance is 3.08% higher than that of the state-of-the-art (SOTA) method.

cs.CL↗
arXiv2026

DEER: Disentangled Mixture of Experts with Instance-Adaptive Routing for Generalizable Machine-Generated Text Detection

Detecting machine-generated text has become a critical challenge amid the rapid advancement of LLMs, yet existing detectors degrade severely under domain shift. Through systematic pilot studies, we trace this vulnerability to two fundamental flaws in current generalization strategies, namely the incomplete preservation of domain-specific knowledge during multi-domain training and the misalignment between knowledge retrieval and the detection objective at inference. To address these gaps, we propose DEER, a Disentangled mixturE-of-ExpeRts framework that explicitly decouples domain-local and domain-invariant knowledge into specialized expert modules. Instead of static domain matching, DEER employs a reinforcement learning-driven router that selects expert pathways based on instance-level detection rewards. This task-aligned, domain-agnostic mechanism ensures robust adaptation to unseen distributions by prioritizing detection utility over stylistic resemblance. Extensive experiments demonstrate that DEER consistently outperforms state-of-the-art detectors, achieving average F1 improvements of 1.28% and 2.92%, and accuracy gains of 1.35% and 2.26% on in-domain and out-of-domain datasets, offering reliable generalization for open-world deployment.

cs.CL↗
arXiv2026

Relativistic Lightsail Propulsion Dynamics

Lightsail technology currently stands as one of the most promising means for interstellar travel, having entered the application phase following multiple successful on-orbit validations. This paper systematically investigates the propulsion effects of incident light, specularly reflected light, and diffusely scattered light on high-speed relativistic lightsails based on radiation dynamics for the first time. The results indicate that, due to the Doppler effect, the thrust from all three components decreases as velocity increases, with incident light providing the strongest thrust, followed by specular reflection, and diffuse scattering being the weakest. Furthermore, the study reveals the existence of a critical velocity for thrust from diffuse scattering: beyond this point, it transforms into drag, although the net radiative force continues to accelerate the spacecraft. By establishing and numerically solving the equations of motion for the lightsail, it is shown that velocity increase is concentrated primarily in the initial acceleration phase, with efficiency declining as speed rises. This research provides a rigorous and in-depth theoretical foundation for lightsail dynamics

astro-ph.IM↗
arXiv2026

SeClaw: Spec-Driven Security Task Synthesis for Evaluating Autonomous Agents

Autonomous LLM agents increasingly operate in stateful environments where they access tools, files, memory, and external services. While such capabilities enable complex real-world workflows, they also introduce security risks that are difficult to capture with existing evaluations. Current agent security benchmarks often rely on manually curated tasks, provide limited coverage of emerging threats, and focus primarily on final outcomes rather than the execution processes that lead to unsafe behavior. We introduce SeClaw, a framework that combines specification-driven security task synthesis with execution-based security evaluation for Autonomous agents. Spec-driven security task synthesis enables scalable and controllable construction of security tasks from structured risk specifications, while SeClaw docker provides a standardized testbed for evaluating agent behavior under diverse safety-risk scenarios. The benchmark covers risks arising from resources, user tasks, environments, and intrinsic agent behaviors, and supports trajectory-aware assessment of unsafe actions beyond final responses. By bridging systematic task synthesis and reproducible security evaluation, SeClaw provides a practical foundation for measuring, diagnosing, and comparing security failures in autonomous LLM agents. The code is available at https://github.com/seclaw-eval/seclaw-eval.

cs.CR↗