arXiv subjects

Yulei Sui

Publications and source records attributed to Yulei Sui.

At least 19 recordsLinked to original sources

arXiv2026

LocalLSTC: A Long Short-Term Control Architecture for Locally Deployed GUI Agents

Modern GUI-agent frameworks achieve strong desktop task performance with frontier API models, yet persistent control information often remains implicit in growing interaction trajectories. At each step, the planner reconstructs the active task stage, accumulated evidence, and runtime feedback before deciding the next action. This dependence becomes more pronounced under weaker local reasoning backbones. Across four representative state-of-the-art frameworks, replacing GPT-5 with Qwen3.5-9B reduces average OSWorld SR-100 from 60.9\% to 37.7\%. Trajectory annotation further identifies at least one control failure in 91.6\% of failed trajectories. To address this problem, we introduce LocalLSTC, a training-free architecture that organizes control by temporal scope, maintaining persistent cross-step state to guide short-term execution commitments. Long-Term Control maintains the active subgoal, subgoal-aligned evidence, and runtime feedback across interactions, while Short-Term Execution realizes bounded commitments for the current step. Long-to-Short Planning forms each commitment from persistent state, and Short-to-Long Control integrates execution outcomes back into that state for progress assessment, recovery, and termination. With Qwen3.6-27B, LocalLSTC reaches 64.7\% SR-100 on OSWorld and 65.3\% on WindowsAgentArena, outperforming the strongest prior local results on both benchmarks. Ablations further support contributions from mechanisms on both sides of execution. These findings identify temporal organization of control information as a distinct architectural dimension for locally deployed GUI agents.

cs.AI↗

arXiv2026

SpreadMark: Robust Image Watermarking via Spread-Spectrum Embedding

Invisible image watermarks are increasingly used for deepfake detection and provenance tracking, where they must survive not only incidental distortions but also deliberate removal. We revisit spread-spectrum embedding, a classical watermarking principle, inside a modern neural post-hoc watermarking architecture. Our starting point is a measurement: in existing encoder-decoder schemes each message bit occupies only a small fraction of the image, a shared contributing factor to their fragility, since removal then need only disturb the region a bit occupies. SpreadMark instead spreads each bit as a dense pseudo-random codeword over the whole image and recovers it by matched-filtering a learned cover-suppressed chip representation, with a parallel convolutional decoding path and sparsification-aware training. A conditional chip-space analysis shows that, under a codeword-independent perturbation model, dense spreading increases the budget required to disrupt matched-filter recovery. Evaluated on COCO and DIV2K against nine schemes, SpreadMark is the only evaluated method retaining high detection under both the regeneration and the latent-space sparsification settings we test, with competitive JPEG and additive-noise robustness. It keeps the embedded watermark imperceptible, maintaining high perceptual quality on both COCO and DIV2K.

cs.CR↗

arXiv2026

Mining Verdict Boundaries for Neural Network Verification

Branch and Bound (BaB) aims to achieve complete verification of neural networks by adaptively partitioning the problem and applying off-the-shelf verifiers to subproblems. Its problem-splitting history can be represented as a tree, where each subproblem corresponds to a child node. A key problem of BaB lies in searching for the verdict boundaries across all the paths that divide the verified and unverified subproblems. We observe that the existing BaB approach tackles this problem by solving each expensive subproblem sequentially along the tree path as its depth increases, requiring costly bounds propagation at every visited BaB tree node (i.e., subproblem), which is inefficient. To address this issue, we propose effective search approaches that leverage the monotonicity of each path to efficiently and precisely locate the verdict boundary by simultaneously splitting multiple activation functions (e.g., ReLU), rather than processing them one at a time as in the classical approach. Our approach performs an effective exponential search along each path, allowing us to skip many boundary-unrelated subproblems when identifying the verdict boundary. The enhanced version further improves this process by estimating the boundary's position using quantitative information obtained from subproblem solving. We perform experimental evaluation on commonly-used benchmarks to assess our proposed techniques, and compare them with recent BaB-based approaches.

cs.LG↗

arXiv2026

Watermark Forensics for Generative Models: An Information-Theoretic Perspective

A watermark in a generative model's output is usually asked only whether a text is machine-made. The same mark can do more: attribute it to the user who produced it, extract a hidden payload, or localize the part that survives editing. These form a forensic ladder, and we ask what each rung costs in the sample length $n$. One object organizes the answers. Let $S$ be the secret the mark carries (a user's identity or payload), and let the information profile $\nu(t)=I(S;X_t\mid X_{<t})$ record how much the $t$-th token reveals about $S$ given the earlier ones. Its total mass pays for attribution and extraction; how that mass is spread pays for localization; and detection alone is paid for not by information but by presence, the distance from the marked to the unmarked distribution. The literature's two quality models, a mark subtle on every token and one that stamps a few tokens loudly, are two incomparable ways of capping this profile. Our main theorem settles the ladder's entropy column. For statistically distortion-free schemes, attributing a text to one of $N$ users costs $\Theta(\log N/h)$ tokens over every stationary-ergodic source of entropy rate $h$, sharp to a $(1+o(1))$ factor: to our knowledge the first tight entropy-rate law for multi-user attribution (via exact alignment). The natural collision-counting analysis overcharges without bound; only a decoder thresholding each candidate by its own realized surprisal attains the rate while almost never implicating an innocent user. A matching converse makes the law two-sided, and extraction of an $\ell$-bit payload costs $\Theta(\ell/h)$. Two gaps are real, not modeling artifacts: a $\Theta(\log N)$-token window in which a text is provably machine-made yet unattributable, and a footprint-resolution uncertainty principle. Experiments on GPT-2, Pythia-410M, and Qwen2.5 recover the predicted constants.

cs.CR↗

arXiv2026

TRACE: A Two-Channel Robust Attribution Watermark via Complementary Embeddings for LLM-Agent Trajectories

LLM agents reach users through resellers, who may rebrand a developer's agent or substitute a cheaper model. When provenance is disputed, attribution rests on the trajectory log (the record of tool calls, observations, and executed actions, not the model's reasoning), which the reseller stores and processes to meter usage. A watermark must therefore survive an adversary with full read/write access to the very evidence it is detected from; existing agent watermarks do not, as their attribution is read straight off that log. We present TRACE, to our knowledge the first agent watermark that is distortion-free in its action choices, self-synchronizing under deletion, and unconditionally invariant under rewriting. Deletion desynchronizes a position-derived key and rewriting alters content, so a deletion-robust key must come from content and a rewrite-robust key from position, and no single key serves both. A trajectory, however, has room for two watermarks. TRACE superposes a selection channel that sets which action is chosen, keyed on local content with a distortion-free sampler, so the agent's distribution is provably unchanged and detection resynchronizes after deletions, and a tally channel that sets how many records each decision group holds, keyed on the log's skeleton alone, which no rewriting can touch. We prove this behavioral watermark's signal is bought with decision entropy, each decision paying at least half its entropy and deterministic decisions nothing, and that erasing both channels forces the reseller to corrupt the trajectories it resells. On ToolBench and ALFWorld, TRACE matches the unwatermarked agent's success rate while its selection channel reaches detection scores near z = 100 on long-horizon trajectories, stays detectable under 70% step deletion, and keeps a tally channel exactly unchanged under LLM rewriting of any strength.

cs.CR↗

arXiv2026

Tensor-Based Batch Fuzzing with Adaptive Perturbation Scaling for Deep Neural Networks

Deep neural networks are increasingly deployed in safety-critical domains such as autonomous driving and medical diagnosis, yet their opaque, high-dimensional parameter spaces make it difficult to systematically assess model reliability on unseen inputs. Existing coverage-guided sequential fuzzing frameworks for DNNs inherit a one-input-per-iteration design from traditional software fuzzing and apply uniform perturbation budgets across all input dimensions, limiting both testing throughput (i.e., inputs processed per unit time) and the precision of input-space exploration. We present a new specification-aware batch fuzzing framework with adaptive perturbation scaling that addresses both limitations. Rather than relying on a fixed global perturbation radius epsilon, our approach derives mutation step sizes from specification-defined feasible ranges (the gap between lower and upper bounds) using a shared scale factor. This scaling can be applied either as a global scalar (isotropic) or as per-dimension step sizes (anisotropic), keeping perturbations consistent with the underlying constraint structure. As a result, the fuzzer can explore input spaces with heterogeneous feature scales more effectively across all specifications in the batch. We embed input constraints and output property checks directly into the network as non-trainable layers, yielding a wrapped model that processes B specification instances in a single batched iteration, substantially improving fuzzing efficiency and counterexample exploration. We evaluate our framework extensively on three benchmarks, covering six networks and over 400 specifications across TrafficSigns, Cifar100, and TinyImageNet. Our tensor-based fuzzing achieves up to 40X higher throughput and 4X more violations than the sequential baseline under the same time budget, demonstrating significantly improved effectiveness in specification-guided fuzzing.

cs.SE↗

arXiv2026

Knowledge Priors for Identity-Disentangled Open-Set Privacy-Preserving Video FER

Facial expression recognition relies on facial data that inherently expose identity and thus raise significant privacy concerns. Current privacy-preserving methods typically fail in realistic open-set video settings where identities are unknown, and identity labels are unavailable. We propose a two-stage framework for video-based privacy-preserving FER in challenging open-set settings that requires no identity labels at any stage. To decouple privacy and utility, we first train an identity-suppression network using intra- and inter-video knowledge priors derived from real-world videos without identity labels. This network anonymizes identity while preserving expressive cues. A subsequent denoising module restores expression-related information and helps recover FER performance. Furthermore, we introduce a falsification-based validation method that uses recognition priors to rigorously evaluate privacy robustness without requiring annotated identity labels. Experiments on three video datasets demonstrate that our method effectively protects privacy while maintaining FER accuracy comparable to identity-supervised baselines.

cs.CV↗

arXiv2026

DeMark: A Query-Free Black-Box Attack on Deepfake Watermarking Defenses

The rapid proliferation of realistic deepfakes has raised urgent concerns over their misuse, motivating the use of defensive watermarks in synthetic images for reliable detection and provenance tracking. However, this defense paradigm assumes such watermarks are inherently resistant to removal. We challenge this assumption with DeMark, a query-free black-box attack framework that targets defensive image watermarking schemes for deepfakes. DeMark exploits latent-space vulnerabilities in encoder-decoder watermarking models through a compressive sensing based sparsification process, suppressing watermark signals while preserving perceptual and structural realism appropriate for deepfakes. Across eight state-of-the-art watermarking schemes, DeMark reduces watermark detection accuracy from 100% to 32.9% on average while maintaining natural visual quality, outperforming existing attacks. We further evaluate three defense strategies, including image super resolution, sparse watermarking, and adversarial training, and find them largely ineffective. These results demonstrate that current encoder decoder watermarking schemes remain vulnerable to latent-space manipulations, underscoring the need for more robust watermarking methods to safeguard against deepfakes.

cs.CR↗

arXiv2025

MetaLogic: Robustness Evaluation of Text-to-Image Models via Logically Equivalent Prompts

Recent advances in text-to-image (T2I) models, especially diffusion-based architectures, have significantly improved the visual quality of generated images. However, these models continue to struggle with a critical limitation: maintaining semantic consistency when input prompts undergo minor linguistic variations. Despite being logically equivalent, such prompt pairs often yield misaligned or semantically inconsistent images, exposing a lack of robustness in reasoning and generalisation. To address this, we propose MetaLogic, a novel evaluation framework that detects T2I misalignment without relying on ground truth images. MetaLogic leverages metamorphic testing, generating image pairs from prompts that differ grammatically but are semantically identical. By directly comparing these image pairs, the framework identifies inconsistencies that signal failures in preserving the intended meaning, effectively diagnosing robustness issues in the model's logic understanding. Unlike existing evaluation methods that compare a generated image to a single prompt, MetaLogic evaluates semantic equivalence between paired images, offering a scalable, ground-truth-free approach to identifying alignment failures. It categorises these alignment errors (e.g., entity omission, duplication, positional misalignment) and surfaces counterexamples that can be used for model debugging and refinement. We evaluate MetaLogic across multiple state-of-the-art T2I models and reveal consistent robustness failures across a range of logical constructs. We find that even the SOTA text-to-image models like Flux.dev and DALLE-3 demonstrate a 59 percent and 71 percent misalignment rate, respectively. Our results show that MetaLogic is not only efficient and scalable, but also effective in uncovering fine-grained logical inconsistencies that are overlooked by existing evaluation metrics.

cs.CV↗

arXiv2025

Efficient Neural Network Verification via Order Leading Exploration of Branch-and-Bound Trees

The vulnerability of neural networks to adversarial perturbations has necessitated formal verification techniques that can rigorously certify the quality of neural networks. As the state-of-the-art, branch and bound (BaB) is a "divide-and-conquer" strategy that applies off-the-shelf verifiers to sub-problems for which they perform better. While BaB can identify the sub-problems that are necessary to be split, it explores the space of these sub-problems in a naive "first-come-first-serve" manner, thereby suffering from an issue of inefficiency to reach a verification conclusion. To bridge this gap, we introduce an order over different sub-problems produced by BaB, concerning with their different likelihoods of containing counterexamples. Based on this order, we propose a novel verification framework Oliva that explores the sub-problem space by prioritizing those sub-problems that are more likely to find counterexamples, in order to efficiently reach the conclusion of the verification. Even if no counterexample can be found in any sub-problem, it only changes the order of visiting different sub-problem and so will not lead to a performance degradation. Specifically, Oliva has two variants, including $Oliva^{GR}$, a greedy strategy that always prioritizes the sub-problems that are more likely to find counterexamples, and $Oliva^{SA}$, a balanced strategy inspired by simulated annealing that gradually shifts from exploration to exploitation to locate the globally optimal sub-problems. We experimentally evaluate the performance of Oliva on 690 verification problems spanning over 5 models with datasets MNIST and CIFAR10. Compared to the state-of-the-art approaches, we demonstrate the speedup of Oliva for up to 25X in MNIST, and up to 80X in CIFAR10.

cs.LG↗

arXiv2025

Tracing Errors, Constructing Fixes: Repository-Level Memory Error Repair via Typestate-Guided Context Retrieval

Memory-related errors in C programming continue to pose significant challenges in software development, primarily due to the complexities of manual memory management inherent in the language. These errors frequently serve as vectors for severe vulnerabilities, while their repair requires extensive knowledge of program logic and C's memory model. Automated Program Repair (APR) has emerged as a critical research area to address these challenges. Traditional APR approaches rely on expert-designed strategies and predefined templates, which are labor-intensive and constrained by the effectiveness of manual specifications. Deep learning techniques offer a promising alternative by automatically extracting repair patterns, but they require substantial training datasets and often lack interpretability. This paper introduces LTFix, a novel approach that harnesses the potential of Large Language Models (LLMs) for automated memory error repair, especially for complex repository-level errors that span multiple functions and files. We address two fundamental challenges in LLM-based memory error repair: a limited understanding of interprocedural memory management patterns and context window limitations for repository-wide analysis. Our approach utilizes a finite typestate automaton to guide the tracking of error-propagation paths and context trace, capturing both spatial (memory states) and temporal (execution history) dimensions of error behavior. This typestate-guided context retrieval strategy provides the LLM with concise yet semantically rich information relevant to erroneous memory management, effectively addressing the token limitation of LLMs.

cs.SE↗

arXiv2025

Adaptive Branch-and-Bound Tree Exploration for Neural Network Verification

Formal verification is a rigorous approach that can provably ensure the quality of neural networks, and to date, Branch and Bound (BaB) is the state-of-the-art that performs verification by splitting the problem as needed and applying off-the-shelf verifiers to sub-problems for improved performance. However, existing BaB may not be efficient, due to its naive way of exploring the space of sub-problems that ignores the \emph{importance} of different sub-problems. To bridge this gap, we first introduce a notion of ``importance'' that reflects how likely a counterexample can be found with a sub-problem, and then we devise a novel verification approach, called ABONN, that explores the sub-problem space of BaB adaptively, in a Monte-Carlo tree search (MCTS) style. The exploration is guided by the ``importance'' of different sub-problems, so it favors the sub-problems that are more likely to find counterexamples. As soon as it finds a counterexample, it can immediately terminate; even though it cannot find, after visiting all the sub-problems, it can still manage to verify the problem. We evaluate ABONN with 552 verification problems from commonly-used datasets and neural network models, and compare it with the state-of-the-art verifiers as baseline approaches. Experimental evaluation shows that ABONN demonstrates speedups of up to $15.2\times$ on MNIST and $24.7\times$ on CIFAR-10. We further study the influences of hyperparameters to the performance of ABONN, and the effectiveness of our adaptive tree exploration.

cs.LG↗

arXiv2025

Towards Source Mapping for Zero-Knowledge Smart Contracts: Design and Preliminary Evaluation

Debugging and auditing zero-knowledge-compatible smart contracts remains a significant challenge due to the lack of source mapping in compilers such as zkSolc. In this work, we present a preliminary source mapping framework that establishes traceability between Solidity source code, LLVM IR, and zkEVM bytecode within the zkSolc compilation pipeline. Our approach addresses the traceability challenges introduced by non-linear transformations and proof-friendly optimizations in zero-knowledge compilation. To improve the reliability of mappings, we incorporate lightweight consistency checks based on static analysis and structural validation. We evaluate the framework on a dataset of 50 benchmark contracts and 500 real-world zkSync contracts, observing a mapping accuracy of approximately 97.2% for standard Solidity constructs. Expected limitations arise in complex scenarios such as inline assembly and deep inheritance hierarchies. The measured compilation overhead remains modest, at approximately 8.6%. Our initial results suggest that source mapping support in zero-knowledge compilation pipelines is feasible and can benefit debugging, auditing, and development workflows. We hope that this work serves as a foundation for further research and tool development aimed at improving developer experience in zk-Rollup environments.

cs.SE↗

arXiv2025

IllusionCAPTCHA: A CAPTCHA based on Visual Illusion

CAPTCHAs have long been essential tools for protecting applications from automated bots. Initially designed as simple questions to distinguish humans from bots, they have become increasingly complex to keep pace with the proliferation of CAPTCHA-cracking techniques employed by malicious actors. However, with the advent of advanced large language models (LLMs), the effectiveness of existing CAPTCHAs is now being undermined. To address this issue, we have conducted an empirical study to evaluate the performance of multimodal LLMs in solving CAPTCHAs and to assess how many attempts human users typically need to pass them. Our findings reveal that while LLMs can solve most CAPTCHAs, they struggle with those requiring complex reasoning type of CAPTCHA that also presents significant challenges for human users. Interestingly, our user study shows that the majority of human participants require a second attempt to pass these reasoning CAPTCHAs, a finding not reported in previous research. Based on empirical findings, we present IllusionCAPTCHA, a novel security mechanism employing the "Human-Easy but AI-Hard" paradigm. This new CAPTCHA employs visual illusions to create tasks that are intuitive for humans but highly confusing for AI models. Furthermore, we developed a structured, step-by-step method that generates misleading options, which particularly guide LLMs towards making incorrect choices and reduce their chances of successfully solving CAPTCHAs. Our evaluation shows that IllusionCAPTCHA can effectively deceive LLMs 100% of the time. Moreover, our structured design significantly increases the likelihood of AI errors when attempting to solve these challenges. Results from our user study indicate that 86.95% of participants successfully passed the CAPTCHA on their first attempt, outperforming other CAPTCHA systems.

cs.CR↗

arXiv2025

CGP-Tuning: Structure-Aware Soft Prompt Tuning for Code Vulnerability Detection

Large language models (LLMs) have been proposed as powerful tools for detecting software vulnerabilities, where task-specific fine-tuning is typically employed to provide vulnerability-specific knowledge to the LLMs. However, existing fine-tuning techniques often treat source code as plain text, losing the graph-based structural information inherent in code. Graph-enhanced soft prompt tuning addresses this by translating the structural information into contextual cues that the LLM can understand. However, current methods are primarily designed for general graph-related tasks and focus more on adjacency information, they fall short in preserving the rich semantic information (e.g., control/data flow) within code graphs. They also fail to ensure computational efficiency while capturing graph-text interactions in their cross-modal alignment module. This paper presents CGP-Tuning, a new code graph-enhanced, structure-aware soft prompt tuning method for vulnerability detection. CGP-Tuning introduces type-aware embeddings to capture the rich semantic information within code graphs, along with an efficient cross-modal alignment module that achieves linear computational costs while incorporating graph-text interactions. It is evaluated on the latest DiverseVul dataset and three advanced open-source code LLMs, CodeLlama, CodeGemma, and Qwen2.5-Coder. Experimental results show that CGP-Tuning delivers model-agnostic improvements and maintains practical inference speed, surpassing the best graph-enhanced soft prompt tuning baseline by an average of four percentage points and outperforming non-tuned zero-shot prompting by 15 percentage points.

cs.SE↗

arXiv2024

Graph Neural Networks for Vulnerability Detection: A Counterfactual Explanation

Vulnerability detection is crucial for ensuring the security and reliability of software systems. Recently, Graph Neural Networks (GNNs) have emerged as a prominent code embedding approach for vulnerability detection, owing to their ability to capture the underlying semantic structure of source code. However, GNNs face significant challenges in explainability due to their inherently black-box nature. To this end, several factual reasoning-based explainers have been proposed. These explainers provide explanations for the predictions made by GNNs by analyzing the key features that contribute to the outcomes. We argue that these factual reasoning-based explanations cannot answer critical what-if questions: What would happen to the GNN's decision if we were to alter the code graph into alternative structures? Inspired by advancements of counterfactual reasoning in artificial intelligence, we propose CFExplainer, a novel counterfactual explainer for GNN-based vulnerability detection. Unlike factual reasoning-based explainers, CFExplainer seeks the minimal perturbation to the input code graph that leads to a change in the prediction, thereby addressing the what-if questions for vulnerability detection. We term this perturbation a counterfactual explanation, which can pinpoint the root causes of the detected vulnerability and furnish valuable insights for developers to undertake appropriate actions for fixing the vulnerability. Extensive experiments on four GNN-based vulnerability detection models demonstrate the effectiveness of CFExplainer over existing state-of-the-art factual reasoning-based explainers.

cs.SE↗

arXiv2024

Iterative Refinement of Project-Level Code Context for Precise Code Generation with Compiler Feedback

Large Language Models (LLMs) have shown remarkable progress in automated code generation. Yet, LLM-generated code may contain errors in API usage, class, data structure, or missing project-specific information. As much of this project-specific context cannot fit into the prompts of LLMs, we must find ways to allow the model to explore the project-level code context. We present CoCoGen, a new code generation approach that uses compiler feedback to improve the LLM-generated code. CoCoGen first leverages static analysis to identify mismatches between the generated code and the project's context. It then iteratively aligns and fixes the identified errors using information extracted from the code repository. We integrate CoCoGen with two representative LLMs, i.e., GPT-3.5-Turbo and Code Llama (13B), and apply it to Python code generation. Experimental results show that CoCoGen significantly improves the vanilla LLMs by over 80% in generating code dependent on the project context and consistently outperforms the existing retrieval-based code generation baselines.

cs.CL↗

arXiv2024

Automated Data Visualization from Natural Language via Large Language Models: An Exploratory Study

The Natural Language to Visualization (NL2Vis) task aims to transform natural-language descriptions into visual representations for a grounded table, enabling users to gain insights from vast amounts of data. Recently, many deep learning-based approaches have been developed for NL2Vis. Despite the considerable efforts made by these approaches, challenges persist in visualizing data sourced from unseen databases or spanning multiple tables. Taking inspiration from the remarkable generation capabilities of Large Language Models (LLMs), this paper conducts an empirical study to evaluate their potential in generating visualizations, and explore the effectiveness of in-context learning prompts for enhancing this task. In particular, we first explore the ways of transforming structured tabular data into sequential text prompts, as to feed them into LLMs and analyze which table content contributes most to the NL2Vis. Our findings suggest that transforming structured tabular data into programs is effective, and it is essential to consider the table schema when formulating prompts. Furthermore, we evaluate two types of LLMs: finetuned models (e.g., T5-Small) and inference-only models (e.g., GPT-3.5), against state-of-the-art methods, using the NL2Vis benchmarks (i.e., nvBench). The experimental results reveal that LLMs outperform baselines, with inference-only models consistently exhibiting performance improvements, at times even surpassing fine-tuned models when provided with certain few-shot demonstrations through in-context learning. Finally, we analyze when the LLMs fail in NL2Vis, and propose to iteratively update the results using strategies such as chain-of-thought, role-playing, and code-interpreter. The experimental results confirm the efficacy of iterative updates and hold great potential for future study.

cs.DB↗