Search arXiv⌕ Search

arXiv subjects

Dehai Zhao

Publications and source records attributed to Dehai Zhao.

At least 19 recordsLinked to original sources
arXiv2026

DualCast: Dual-Image Conditioned Diffusion Transformers for Digital Human E-Commerce Video Generation

Digital human e-commerce video generation requires preserving product appearance, human identity, and overall video style during dynamic demonstrations. Although recent models support multiple reference images, their general conditioning schemes do not explicitly distinguish the asymmetric roles of "what is presented" and "who presents it." Consequently, generated videos may favor a single reference, resulting in product distortion, identity drift, or single-reference collapse. To address this issue, we propose DualCast, a role-addressed dual-image conditioning framework for video generation. By explicitly modeling the distinct roles of product and human conditions, DualCast reduces role ambiguity and feature competition between references. First, a dual-spatiotemporal slot injection mechanism VAE-encodes the product and human images separately and places their latent features at the front and rear boundaries of the DiT sequence, enabling sequence-level condition separation and directed injection. Second, a mask-guided region-weighted loss strengthens supervision over product, face, and product-face overlap regions, improving generation fidelity in critical areas. Third, image condition dropout enables a single model to support dual-image, product-only, and human-only generation. In addition, we construct a curated dataset of 10K e-commerce videos for model training. Experimental results demonstrate the strong performance of DualCast.

cs.CV↗
arXiv2026

CrossPair: Unlocking Generalist Visual Transformation Learning in Image Editors

Language models switch tasks from demonstrations, but vision still often relies on separate systems for restoration, editing, and stylization. A before--after example offers a shared instruction only if the transferable change can be separated from the scene that demonstrates it. We introduce CrossPair, which adapts a pretrained multimodal image editor to transfer diverse transformations from one exemplar pair. Exemplar-before, exemplar-after, and query images take fixed roles in its native VLM and VAE pathways: the VLM extracts operation semantics, while the VAE retains fine appearance and structure. A training-only, change-aware DINOv3 objective aligns feature-change directions across content-distinct scenes, guiding the output to follow the demonstrated relation without an inference-time encoder. We construct Transform-360K by grouping filtered pairs from conditional representations, restoration, editing, virtual try-on, and stylization under shared transformations. On Relation252K-derived seen and held-out transformations, \method improves target agreement and edit-direction similarity over released systems. Ablations show that the semantic and visual exemplar routes are complementary and that DINO relation alignment further strengthens transfer; Transform-360K training also improves aggregate held-out-operation scores.

cs.CV↗
arXiv2026

One Evaluation Is Enough: Inversion-Free Image Editing with Trajectory Correction

Real-image editing commonly relies on inversion and iterative sampling, while efficient one-step editors often stabilize a large update by estimating and averaging editing fields at multiple noise levels. We ask a stricter question: is a single temporal field evaluation sufficient for reliable inversion-free editing? Our answer rests on a previously underused duality in a variance-preserving (VP) one-step model: one source-target query at a shared noisy state simultaneously determines a local angular semantic tangent and a clean endpoint displacement. We integrate the tangent over the analytically known remaining VP angle to obtain a nominal one-step transport, and use the endpoint displacement to correct its long-horizon extrapolation through a constrained projection. This transport-first, correction-second construction yields Endpoint-Calibrated Transport (ECT), a closed-form, training-free editor requiring only one noise level, two text conditions packed into one denoiser batch, and one latent update. ECT requires no image inversion, trajectory integration, second temporal observation, or iterative optimization, and can be directly applied to pretrained epsilon-prediction VP one-step generators. Across the reported benchmark settings, ECT delivers stable gains in editing quality within an extremely short editing time.

cs.CV↗
arXiv2026

CAVC: Convergence-Adaptive Audio-Visual Cache for Modality-Asymmetric Diffusion Generation

Diffusion-based audio-visual generation methods typically perform joint denoising of audio and video latents throughout the entire sampling trajectory, repeatedly executing modality-specific denoising branches and cross-modal attention at every timestep. However, we observe that such uniform joint computation is not always necessary. The key is a modality-asymmetric recoverability: audio-side drift introduced by mid-stage caching can be substantially reduced by only a few terminal joint denoising steps, whereas video-side degradation caused by caching leads to persistent loss of motion and temporal details. This asymmetry suggests that inference acceleration should selectively cache recoverable audio-side computation while preserving full video refinement. Motivated by this observation, we propose \textbf{Convergence-Adaptive Audio-Visual Cache (CAVC)}, a training-free modality-asymmetric caching framework. CAVC adopts a sandwich-structured computation allocation: after detecting that audio denoiser outputs enter a cache-tolerant regime, it caches and reuses the audio denoiser output to bypass audio denoising and cross-modal attention in the middle stage, while the video branch continues independent spatio-temporal refinement; a few joint steps are then restored at the end to reduce accumulated cross-modal drift and improve synchronization. Experiments across multiple public benchmarks demonstrate that CAVC achieves 1.10--1.35x wall-clock acceleration, improves several visual dynamics and temporal statistics, and maintains competitive audio fidelity and audio-visual alignment.

cs.CV↗
arXiv2026

WorldWeave: Growing Persistent Geometric Worlds for Video Generation

Despite rapid progress, world models still lack explicit, persistent structural memory, making it difficult to preserve consistent world structure during continual scene expansion and cross-view revisits. To address this limitation, we present WorldWeave, a world generation framework that decouples world-state maintenance from visual rendering. Specifically, WorldWeave combines continual elevation-map generation with agent-guided scene organization and stitching to build an expandable explicit 3D world that incrementally extends structural memory while preserving existing structure. First, its terrain module uses diffusion-based image outpainting to generate continuous metric elevation maps under neighborhood conditioning and boundary constraints. Next, an agent integrates user intent, terrain evidence, and cross-region connectivity constraints to construct scenes through hierarchical semantic planning, deterministic geometry compilation, and local revision. Finally, during visual generation, planned camera trajectories query world geometry through a read-only interface, producing depth sequences that guide video synthesis without writing the generated results back into the world state. As a result, structural memory remains independent of short-window video generation, enabling continual expansion without predefined map boundaries and providing a consistent geometric basis for observations across trajectories and repeated visits.

cs.CV↗
arXiv2026

An Empirical Study of Agent Skills' Downstream Utility

Agent Skills package procedural guidance and resources for reuse, but a relevant Skill does not necessarily improve task performance. Existing studies characterize Skill content and evaluate downstream performance, yet provide limited explanations of how utility depends on content, execution configuration, and multi-Skill organization. We conduct an empirical study on 87 SkillsBench tasks, defining downstream utility as the pass-rate difference from No-Skill on the same tasks under the same model--harness configuration. We compare the same Skills across nine configurations, then examine alternative published Skills and organizations of fixed Skill sets under three selected configurations. We retrieve marketplace candidates from a curated corpus of 37,596 Skills. LLM-assisted analysis of content, execution traces, and final artifacts, followed by author review, relates provided support to actual use and task outcomes. The same Skills help some configurations and hurt others on 36.78\% of tasks, with trajectories showing that recommended procedures can become an execution burden. Relevance rankings overlook more useful candidates. Within the evaluated candidate sets, reranking by support for required operations raises first-choice pass rates by 4.35--5.80 percentage points across the three configurations. We derive 17 authoring practices linking executable procedures to recovery, preservation of task requirements, and checks on final artifacts. Stage Plan and Dependency DAG outperform use order alone, with DAG's additional benefits concentrated in tasks supplied with five or six Skills. These findings guide developers to assess usable operation support, allow procedure adaptation while preserving task requirements, and make artifact dependencies explicit when organizing Skills.

cs.AI↗
arXiv2026

DeGRe: Dense-supervised Generative Reranking for Recommendation

In multi-stage recommender systems, reranking optimizes overall utility by capturing intra-list contextual dependencies, yet its central challenge lies in exploring optimal sequences within an exponentially large permutation space. Recent studies have shifted towards end-to-end generative frameworks, which typically leverage list-wise rewards or preference alignment to guide generator training. However, these methods still face two critical issues. First is the heuristic label bias. Existing methods often construct training targets based on simple rules, such as promoting clicked items to the top, while ignoring causal dependencies within the list context. Second is the credit assignment problem. Sparse list-level posterior rewards fail to directly guide intermediate steps in sequence generation, leading to ambiguous optimization directions. To address these issues, we propose DeGRe (Dense-supervised Generative Reranking), a generative reranking framework that bridges the gap between offline exploration and online efficiency through dense supervision. The core of DeGRe lies in its offline-online decoupled design. During the offline phase, we introduce a Lookahead Evaluator based on cumulative regression, which leverages beam search to actively mine high-value lookahead sequences in the unexposed space. During training, we transform the step-wise value estimations from the evaluator into dense supervision signals and distill them into a lightweight Online Generator. This mechanism enables the generator to internalize lookahead planning capabilities, requiring only a single efficient greedy decoding pass during online inference to approximate the global optimum. Experiments demonstrate that DeGRe outperforms baseline models on public benchmarks and industrial datasets. We have successfully deployed DeGRe on Taobao Flash Shopping, significantly improving online recommendations.

cs.IR↗
arXiv2025

Evaluation-Driven Development and Operations of LLM Agents: A Process Model and Reference Architecture

Large Language Models (LLMs) have enabled the emergence of LLM agents, systems capable of pursuing under-specified goals and adapting after deployment. Evaluating such agents is challenging because their behavior is open ended, probabilistic, and shaped by system-level interactions over time. Traditional evaluation methods, built around fixed benchmarks and static test suites, fail to capture emergent behaviors or support continuous adaptation across the lifecycle. To ground a more systematic approach, we conduct a multivocal literature review (MLR) synthesizing academic and industrial evaluation practices. The findings directly inform two empirically derived artifacts: a process model and a reference architecture that embed evaluation as a continuous, governing function rather than a terminal checkpoint. Together they constitute the evaluation-driven development and operations (EDDOps) approach, which unifies offline (development-time) and online (runtime) evaluation within a closed feedback loop. By making evaluation evidence drive both runtime adaptation and governed redevelopment, EDDOps supports safer, more traceable evolution of LLM agents aligned with changing objectives, user needs, and governance constraints.

cs.SE↗
arXiv2025

AgentArcEval: An Architecture Evaluation Method for Foundation Model based Agents

The emergence of foundation models (FMs) has enabled the development of highly capable and autonomous agents, unlocking new application opportunities across a wide range of domains. Evaluating the architecture of agents is particularly important as the architectural decisions significantly impact the quality attributes of agents given their unique characteristics, including compound architecture, autonomous and non-deterministic behaviour, and continuous evolution. However, these traditional methods fall short in addressing the evaluation needs of agent architecture due to the unique characteristics of these agents. Therefore, in this paper, we present AgentArcEval, a novel agent architecture evaluation method designed specially to address the complexities of FM-based agent architecture and its evaluation. Moreover, we present a catalogue of agent-specific general scenarios, which serves as a guide for generating concrete scenarios to design and evaluate the agent architecture. We demonstrate the usefulness of AgentArcEval and the catalogue through a case study on the architecture evaluation of a real-world tax copilot, named Luna.

cs.SE↗
arXiv2025

Bridging Solidity Evolution Gaps: An LLM-Enhanced Approach for Smart Contract Compilation Error Resolution

Solidity, the dominant smart contract language for Ethereum, has rapidly evolved with frequent version updates to enhance security, functionality, and developer experience. However, these continual changes introduce significant challenges, particularly in compilation errors, code migration, and maintenance. Therefore, we conduct an empirical study to investigate the challenges in the Solidity version evolution and reveal that 81.68% of examined contracts encounter errors when compiled across different versions, with 86.92% of compilation errors. To mitigate these challenges, we conducted a systematic evaluation of large language models (LLMs) for resolving Solidity compilation errors during version migrations. Our empirical analysis across both open-source (LLaMA3, DeepSeek) and closed-source (GPT-4o, GPT-3.5-turbo) LLMs reveals that although these models exhibit error repair capabilities, their effectiveness diminishes significantly for semantic-level issues and shows strong dependency on prompt engineering strategies. This underscores the critical need for domain-specific adaptation in developing reliable LLM-based repair systems for smart contracts. Building upon these insights, we introduce SMCFIXER, a novel framework that systematically integrates expert knowledge retrieval with LLM-based repair mechanisms for Solidity compilation error resolution. The architecture comprises three core phases: (1) context-aware code slicing that extracts relevant error information; (2) expert knowledge retrieval from official documentation; and (3) iterative patch generation for Solidity migration. Experimental validation across Solidity version migrations demonstrates our approach's statistically significant 24.24% improvement over baseline GPT-4o on real-world datasets, achieving near-perfect 96.97% accuracy.

cs.SE↗
arXiv2025

SHIELDA: Structured Handling of Exceptions in LLM-Driven Agentic Workflows

Large Language Model (LLM) agentic systems are software systems powered by LLMs that autonomously reason, plan, and execute multi-step workflows to achieve human goals, rather than merely executing predefined steps. During execution, these workflows frequently encounter exceptions. Existing exception handling solutions often treat exceptions superficially, failing to trace execution-phase exceptions to their reasoning-phase root causes. Furthermore, their recovery logic is brittle, lacking structured escalation pathways when initial attempts fail. To tackle these challenges, we first present a comprehensive taxonomy of 36 exception types across 12 agent artifacts. Building on this, we propose SHIELDA (Structured Handling of Exceptions in LLM-Driven Agentic Workflows), a modular runtime exception handling framework for LLM agentic workflows. SHIELDA uses an exception classifier to select a predefined exception handling pattern from a handling pattern registry. These patterns are then executed via a structured handling executor, comprising local handling, flow control, and state recovery, to enable phase-aware recovery by linking exceptions to their root causes and facilitating composable strategies. We validate SHIELDA's effectiveness through a case study on the AutoPR agent, demonstrating effective, cross-phase recovery from a reasoning-induced exception.

cs.SE↗
arXiv2025

When Prompt Engineering Meets Software Engineering: CNL-P as Natural and Robust "APIs'' for Human-AI Interaction

With the growing capabilities of large language models (LLMs), they are increasingly applied in areas like intelligent customer service, code generation, and knowledge management. Natural language (NL) prompts act as the ``APIs'' for human-LLM interaction. To improve prompt quality, best practices for prompt engineering (PE) have been developed, including writing guidelines and templates. Building on this, we propose Controlled NL for Prompt (CNL-P), which not only incorporates PE best practices but also draws on key principles from software engineering (SE). CNL-P introduces precise grammar structures and strict semantic norms, further eliminating NL's ambiguity, allowing for a declarative but structured and accurate expression of user intent. This helps LLMs better interpret and execute the prompts, leading to more consistent and higher-quality outputs. We also introduce an NL2CNL-P conversion tool based on LLMs, enabling users to write prompts in NL, which are then transformed into CNL-P format, thus lowering the learning curve of CNL-P. In particular, we develop a linting tool that checks CNL-P prompts for syntactic and semantic accuracy, applying static analysis techniques to NL for the first time. Extensive experiments demonstrate that CNL-P enhances the quality of LLM responses through the novel and organic synergy of PE and SE. We believe that CNL-P can bridge the gap between emerging PE and traditional SE, laying the foundation for a new programming paradigm centered around NL.

cs.SE↗
arXiv2025

Think Like an Engineer: A Neuro-Symbolic Collaboration Agent for Generative Software Requirements Elicitation and Self-Review

The vision of End-User Software Engineering (EUSE) is to empower non-professional users with full control over the software development lifecycle. It aims to enable users to drive generative software development using only natural language requirements. However, since end-users often lack knowledge of software engineering, their requirement descriptions are frequently ambiguous, raising significant challenges to generative software development. Although existing approaches utilize structured languages like Gherkin to clarify user narratives, they still struggle to express the causal logic between preconditions and behavior actions. This paper introduces RequireCEG, a requirement elicitation and self-review agent that embeds causal-effect graphs (CEGs) in a neuro-symbolic collaboration architecture. RequireCEG first uses a feature tree to analyze user narratives hierarchically, clearly defining the scope of software components and their system behavior requirements. Next, it constructs the self-healing CEGs based on the elicited requirements, capturing the causal relationships between atomic preconditions and behavioral actions. Finally, the constructed CEGs are used to review and optimize Gherkin scenarios, ensuring consistency between the generated Gherkin requirements and the system behavior requirements elicited from user narratives. To evaluate our method, we created the RGPair benchmark dataset and conducted extensive experiments. It achieves an 87% coverage rate and raises diversity by 51.88%.

cs.SE↗
arXiv2025

Privacy Meets Explainability: Managing Confidential Data and Transparency Policies in LLM-Empowered Science

As Large Language Models (LLMs) become integral to scientific workflows, concerns over the confidentiality and ethical handling of confidential data have emerged. This paper explores data exposure risks through LLM-powered scientific tools, which can inadvertently leak confidential information, including intellectual property and proprietary data, from scientists' perspectives. We propose "DataShield", a framework designed to detect confidential data leaks, summarize privacy policies, and visualize data flow, ensuring alignment with organizational policies and procedures. Our approach aims to inform scientists about data handling practices, enabling them to make informed decisions and protect sensitive information. Ongoing user studies with scientists are underway to evaluate the framework's usability, trustworthiness, and effectiveness in tackling real-world privacy challenges.

cs.HC↗
arXiv2025

SeeAction: Towards Reverse Engineering How-What-Where of HCI Actions from Screencasts for UI Automation

UI automation is a useful technique for UI testing, bug reproduction, and robotic process automation. Recording user actions with an application assists rapid development of UI automation scripts, but existing recording techniques are intrusive, rely on OS or GUI framework accessibility support, or assume specific app implementations. Reverse engineering user actions from screencasts is non-intrusive, but a key reverse-engineering step is currently missing - recognizing human-understandable structured user actions ([command] [widget] [location]) from action screencasts. To fill the gap, we propose a deep learning-based computer vision model that can recognize 11 commands and 11 widgets, and generate location phrases from action screencasts, through joint learning and multi-task learning. We label a large dataset with 7260 video-action pairs, which record user interactions with Word, Zoom, Firefox, Photoshop, and Windows 10 Settings. Through extensive experiments, we confirm the effectiveness and generality of our model, and demonstrate the usefulness of a screencast-to-action-script tool built upon our model for bug reproduction.

cs.SE↗
arXiv2025

Swiss Cheese Model for AI Safety: A Taxonomy and Reference Architecture for Multi-Layered Guardrails of Foundation Model Based Agents

Foundation Model (FM)-based agents are revolutionizing application development across various domains. However, their rapidly growing capabilities and autonomy have raised significant concerns about AI safety. Researchers are exploring better ways to design guardrails to ensure that the runtime behavior of FM-based agents remains within specific boundaries. Nevertheless, designing effective runtime guardrails is challenging due to the agents' autonomous and non-deterministic behavior. The involvement of multiple pipeline stages and agent artifacts, such as goals, plans, tools, at runtime further complicates these issues. Addressing these challenges at runtime requires multi-layered guardrails that operate effectively at various levels of the agent architecture. Therefore, in this paper, based on the results of a systematic literature review, we present a comprehensive taxonomy of runtime guardrails for FM-based agents to identify the key quality attributes for guardrails and design dimensions. Inspired by the Swiss Cheese Model, we also propose a reference architecture for designing multi-layered runtime guardrails for FM-based agents, which includes three dimensions: quality attributes, pipelines, and artifacts. The proposed taxonomy and reference architecture provide concrete and robust guidance for researchers and practitioners to build AI-safety-by-design from a software architecture perspective.

cs.SE↗
arXiv2024

Agent Design Pattern Catalogue: A Collection of Architectural Patterns for Foundation Model based Agents

Foundation model-enabled generative artificial intelligence facilitates the development and implementation of agents, which can leverage distinguished reasoning and language processing capabilities to takes a proactive, autonomous role to pursue users' goals. Nevertheless, there is a lack of systematic knowledge to guide practitioners in designing the agents considering challenges of goal-seeking (including generating instrumental goals and plans), such as hallucinations inherent in foundation models, explainability of reasoning process, complex accountability, etc. To address this issue, we have performed a systematic literature review to understand the state-of-the-art foundation model-based agents and the broader ecosystem. In this paper, we present a pattern catalogue consisting of 18 architectural patterns with analyses of the context, forces, and trade-offs as the outcomes from the previous literature review. We propose a decision model for selecting the patterns. The proposed catalogue can provide holistic guidance for the effective use of patterns, and support the architecture design of foundation model-based agents by facilitating goal-seeking and plan generation.

cs.AI↗
arXiv2024

CasModaTest: A Cascaded and Model-agnostic Self-directed Framework for Unit Test Generation

Though many machine learning (ML)-based unit testing generation approaches have been proposed and indeed achieved remarkable performance, they still have several limitations in effectiveness and practical usage. More precisely, existing ML-based approaches (1) generate partial content of a unit test, mainly focusing on test oracle generation; (2) mismatch the test prefix with the test oracle semantically; and (3) are highly bound with the close-sourced model, eventually damaging data security. We propose CasModaTest, a cascaded, model-agnostic, and end-to-end unit test generation framework, to alleviate the above limitations with two cascaded stages: test prefix generation and test oracle generation. Then, we manually build large-scale demo pools to provide CasModaTest with high-quality test prefixes and test oracles examples. Finally, CasModaTest automatically assembles the generated test prefixes and test oracles and compiles or executes them to check their effectiveness, optionally appending with several attempts to fix the errors occurring in compiling and executing phases. To evaluate the effectiveness of CasModaTest, we conduct large-scale experiments on a widely used dataset (Defects4J) and compare it with four state-of-the-art (SOTA) approaches by considering two performance measures. The experimental results indicate that CasModaTest outperforms all SOTAs with a substantial improvement (i.e., 60.62%-352.55% in terms of accuracy, 2.83%-87.27% in terms of focal method coverage). Besides, we also conduct experiments of CasModaTest on different open-source LLMs and find that CasModaTest can also achieve significant improvements over SOTAs (39.82%-293.96% and 9.25%-98.95% in terms of accuracy and focal method coverage, respectively) in end-to-end unit test generation

cs.SE↗