arXiv subjects

Miao Yu

Publications and source records attributed to Miao Yu.

At least 19 recordsLinked to original sources

arXiv2026

Training-Free Logical and Structural Anomaly Detection via Calibrated Fusion

Industrial anomaly detection must handle two distinct defect families: structural anomalies, which manifest as local texture corruptions, and logical anomalies, which violate global rules on object count, composition, or arrangement. Existing detectors typically favor one family at the expense of the other. In particular, training-free methods effectively exploit frozen representations but lack an explicit notion of object count, while methods that reason about counts usually rely on category-specific component modeling. We show that counting ability can be introduced into training-free anomaly detection without additional training or part-level supervision. Our key idea is a normal-set calibration that aligns heterogeneous anomaly cues using statistics from normal images, enabling their direct fusion within a unified training-free framework. Built upon this calibration, our detector combines complementary frozen cues to address both logical and structural anomalies. On MVTec-LOCO, our method achieves image-level AUROCs of 89.0 and 95.9 on logical and structural anomalies, respectively, yielding a 92.5 average---the best among training-free detectors in our comparison. It remains competitive with methods requiring network training or part annotations, while its structural variant matches PatchCore on MVTec-AD (99.1 image-AUROC), suggesting that the proposed calibration generalizes beyond logical anomaly detection.

cs.CV↗

arXiv2026

CAVE-NAV: VLM-Based Autonomous 3D Navigation in Underwater Cave Environments

Autonomous navigation in underwater cave environments is essential for search-and-rescue operations, scientific exploration, and emergency egress. Traditional navigation systems commonly depend on dense visual features for localization and mapping. In underwater caves, however, visual degradation can undermine feature-based localization, sonar-based mapping may yield overly conservative obstacle representations, and communication constraints preclude real-time human guidance. To address these limitations, we propose an autonomous underwater cave navigation framework that leverages a vision-language model (VLM) with Chain-of-Thought (CoT) reasoning to infer navigable directions from environmental cues, including light intensity gradients, passage morphology, and geometric complexity, captured through multimodal inputs comprising RGB imagery, depth maps, and sonar-based vertical-clearance measurements, thereby supporting safe 3D navigation through confined cave passages. High-fidelity simulations across multiple cave topologies demonstrate that the proposed framework completes all evaluated end-to-end traversals without collisions while maintaining safe clearance from cave boundaries.

cs.RO↗

arXiv2026

A Trajectory-Driven Spatio-Temporal Refinement Solution for CVPR 2026 8th UG2+ Challenge Track 3: DOST

In this work, we present our solution for the 8th UG2+ Challenge (CVPR 2026) Track 3: Dynamic Object Segmentation in Turbulence (DOST). Our method is built upon the strong baseline framework Segment Any Motion (SegAnyMo), which provides powerful mask generation and motion tracking capabilities. To further boost the segmentation performance under severe atmospheric distortions, we propose two key improvements. First, we employ a data-centric domain adaptation strategy. We significantly expand our training data by incorporating selected sequences from the DAVIS dataset alongside a subset of the DOST dataset, and apply simulated atmospheric fluctuation degradations to enhance the model's robustness against complex geometric distortions. Second, we introduce a spatio-temporal post-processing module. This refinement step effectively removes persistent boundary-connected false foregrounds and short-lived fragmented noise, while strictly preserving genuine small targets and maintaining original individual labels across frames. With these combined strategies, our proposed method ranks the 2st place in the challenge.

cs.CV↗

arXiv2026

A Survey of Large Audio Language Models: Generalization, Trustworthiness, and Outlook

Advances in Large Language Models (LLMs) have paved the way for Multimodal Large Language Models (MLLMs). Among these, Large Audio Language Models (LALMs) are essential for realizing universal auditory intelligence. Despite their remarkable performance, the escalation of LALMs' capabilities has significantly outpaced the development of systemic frameworks to ensure their trustworthiness. This survey provides a comprehensive investigation into the endogenous mechanisms of LALMs, detailing the architectural innovations and alignment algorithms that facilitate emergent reasoning. Specifically, we analyze how the transition to unified end-to-end frameworks and the integration of continuous acoustic signals expand the attack surface. To rigorously evaluate the risks within these paradigms, we establish a comprehensive taxonomy of trustworthiness, categorizing critical vulnerabilities such as cross-modal jailbreaking, latent acoustic backdoors, and biometric privacy leakage. We review the state-of-the-art LALMs through six analytical pillars: hallucination, robustness, safety, privacy, fairness, and authentication. The pronounced imbalance between a mature offensive landscape and underdeveloped defenses highlights persistent trustworthiness gaps and multidimensional risks in audio-centric intelligence. Finally, we propose a roadmap advocating for ``Defense-in-Depth'' architectures, causal auditory world modeling, and intrinsic representation engineering to support the development of more reliable and trustworthy audio intelligence. Our project has been uploaded to GitHub https://github.com/Kwwwww74/Awesome-Trustworthy-AudioLLMs.

cs.SD↗

arXiv2026

Sonar-GPS Fusion for Seabed Mapping in Turbid Shallow Waters with an Autonomous Surface Vehicle

Accurate seabed mapping is essential for habitat monitoring and infrastructure inspection. In turbid, shallow coastal waters, such as shellfish aquaculture farms, the effectiveness of traditional optical methods is limited. Autonomous surface vehicles (ASVs) equipped with forward-looking sonar (FLS) offer a promising alternative. However, existing sonar-based systems face challenges in achieving fine resolution mapping over long trajectories due to low-resolution positioning measurements and accumulated drift over long trajectories. In this paper, we present a drift-resilient seabed mapping framework that integrates local FLS frame alignment using the Fourier-Mellin transform (FMT) with global trajectory optimization based on an extended Kalman filter (EKF) that fuses global positioning system (GPS), inertial measurement unit (IMU), and compass data. A variance-based image blending strategy is used to further reduce visual artifacts in overlapping regions. Field trials on a structured oyster farm site show that our framework helps reduce drift in RMSE by 9.5% relative to the FMT-only baseline. This framework also enables sub-meter reconstruction accuracy and preservation of high-resolution textures needed for oyster inventory estimation within the mapped areas.

cs.RO↗

arXiv2026

ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety

Multimodal Large Language Models (MLLMs) have achieved remarkable success in cross-modal understanding and generation, yet their deployment is threatened by critical safety vulnerabilities. While prior works have demonstrated the feasibility of backdoors in MLLMs via fine-tuning data poisoning to manipulate inference, the underlying mechanisms of backdoor attacks remain opaque, complicating the understanding and mitigation. To bridge this gap, we propose ProjLens, an interpretability framework designed to demystify MLLMs backdoors. We first establish that normal downstream task alignment--even when restricted to projector fine--tuning--introduces vulnerability to backdoor injection, whose activation mechanism is different from that observed in text-only LLMs. Through extensive experiments across four backdoor variants, we uncover:(1) Low-Rank Structure: Backdoor injection updates appear overall full-rank and lack dedicated ``trigger neurons'', but the backdoor-critical parameters are encoded within a low-rank subspace of the projector;(2) Activation Mechanism: Both clean and poisoned embedding undergoes a semantic shift toward a shared direction aligned with the backdoor target, but the shifting magnitude scales linearly with the input norm, resulting in the distinct backdoor activation on poisoned samples. Our code is available at: https://anonymous.4open.science/r/ProjLens-8FD7

cs.CR↗

arXiv2026

SafeSeek: Universal Attribution of Safety Circuits in Language Models

Mechanistic interpretability reveals that safety-critical behaviors (e.g., alignment, jailbreak, backdoor) in Large Language Models (LLMs) are grounded in specialized functional components. However, existing safety attribution methods struggle with generalization and reliability due to their reliance on heuristic, domain-specific metrics and search algorithms. To address this, we propose \ourmethod, a unified safety interpretability framework that identifies functionally complete safety circuits in LLMs via optimization. Unlike methods focusing on isolated heads or neurons, \ourmethod introduces differentiable binary masks to extract multi-granular circuits through gradient descent on safety datasets, while integrates Safety Circuit Tuning to utilize these sparse circuits for efficient safety fine-tuning. We validate \ourmethod in two key scenarios in LLM safety: \textbf{(1) backdoor attacks}, identifying a backdoor circuit with 0.42\% sparsity, whose ablation eradicates the Attack Success Rate (ASR) from 100\% $\to$ 0.4\% while retaining over 99\% general utility; \textbf{(2) safety alignment}, localizing an alignment circuit with 3.03\% heads and 0.79\% neurons, whose removal spikes ASR from 0.8\% $\to$ 96.9\%, whereas excluding this circuit during helpfulness fine-tuning maintains 96.5\% safety retention.

cs.LG↗

arXiv2026

MBD: A Model-Based Debiasing Framework Across User, Content, and Model Dimensions

Modern recommendation systems rank candidates by aggregating multiple behavioral signals through a value model. However, many commonly used signals are inherently affected by heterogeneous biases. For example, watch time naturally favors long-form content, loop rate favors short - form content, and comment probability favors videos over images. Such biases introduce two critical issues: (1) value model scores may be systematically misaligned with users' relative preferences - for instance, a seemingly low absolute like probability may represent exceptionally strong interest for a user who rarely engages; and (2) changes in value modeling rules can trigger abrupt and undesirable ecosystem shifts. In this work, we ask a fundamental question: can biased behavioral signals be systematically transformed into unbiased signals, under a user - defined notion of ``unbiasedness'', that are both personalized and adaptive? We propose a general, model-based debiasing (MBD) framework that addresses this challenge by augmenting it with distributional modeling. By conditioning on a flexible subset of features (partial feature set), we explicitly estimate the contextual mean and variance of the engagement distribution for arbitrary cohorts (e.g., specific video lengths or user regions) directly alongside the main prediction. This integration allows the framework to convert biased raw signals into unbiased representations, enabling the construction of higher-level, calibrated signals (such as percentiles or z - scores) suitable for the value model. Importantly, the definition of unbiasedness is flexible and controllable, allowing the system to adapt to different personalization objectives and modeling preferences. Crucially, this is implemented as a lightweight, built-in branch of the existing MTML ranking model, requiring no separate serving infrastructure.

cs.LG↗

arXiv2026

Long-term Monitoring of Kernel and Hardware Events to Understand Latency Variance

This paper presents our experience to understand latency variance caused by kernel and hardware events, which are often invisible at the application level. For this purpose, we have built VarMRI, a tool chain to monitor and analyze those events in the long term. To mitigate the "big data" problem caused by long-term monitoring, VarMRI selectively records a subset of events following two principles: it only records events that are affecting the requests recorded by the application; it records coarse-grained information first and records additional information only when necessary. Furthermore, VarMRI introduces an analysis method that is efficient on large amount of data, robust on different data set and against missing data, and informative to the user. VarMRI has helped us to carry out a 3,000-hour study of six applications and benchmarks on CloudLab. It reveals a wide variety of events causing latency variance, including interrupt preemption, Java GC, pipeline stall, NUMA balancing etc.; simple optimization or tuning can reduce tail latencies by up to 31%. Furthermore, the impacts of some of these events vary significantly across different experiments, which confirms the necessity of long-term monitoring.

cs.PF↗

arXiv2025

Neural Architecture Search of Time-to-First-Spike-Coded Spiking Neural Networks for Efficient Eye-based Emotion Recognition

Eye-based emotion recognition enables eyewear devices to perceive users' emotional states and support emotion-aware interaction. However, deploying such functionality on their resource-limited embedded hardware remains challenging. Time-to-first-spike (TTFS)-coded spiking neural networks (SNNs) offer a promising solution due to their extremely sparse and energy-efficient computation, where each neuron emits at most one binary spike. While prior works have primarily focused on improving TTFS SNN training algorithms, the role of network architecture has been largely overlooked. This is particularly critical, as spike timing in TTFS SNNs is tightly coupled with architectural design, and eye-based emotion recognition requires compact yet highly efficient networks. In this paper, we propose TNAS-ER, the first neural architecture search (NAS) framework tailored to TTFS SNNs for eye-based emotion recognition. TNAS-ER presents a novel ANN-assisted search strategy that leverages a ReLU-based ANN counterpart to guide architecture optimization and stabilize training of the TTFS SNN. TNAS-ER employs an evolutionary algorithm, with weighted and unweighted average recall jointly defined as fitness objectives for emotion recognition. Extensive experiments demonstrate that TNAS-ER achieves high recognition performance with significantly improved efficiency. Furthermore, we evaluate TNAS-ER on a neuromorphic hardware, confirming its superior energy efficiency and strong potential for real-world applications.

cs.NE↗

arXiv2025

EmoRAG: Evaluating RAG Robustness to Symbolic Perturbations

Retrieval-Augmented Generation (RAG) systems are increasingly central to robust AI, enhancing large language model (LLM) faithfulness by incorporating external knowledge. However, our study unveils a critical, overlooked vulnerability: their profound susceptibility to subtle symbolic perturbations, particularly through near-imperceptible emoticon tokens such as "(@_@)" that can catastrophically mislead retrieval, termed EmoRAG. We demonstrate that injecting a single emoticon into a query makes it nearly 100% likely to retrieve semantically unrelated texts that contain a matching emoticon. Our extensive experiment across general question-answering and code domains, using a range of state-of-the-art retrievers and generators, reveals three key findings: (I) Single-Emoticon Disaster: Minimal emoticon injections cause maximal disruptions, with a single emoticon almost 100% dominating RAG output. (II) Positional Sensitivity: Placing an emoticon at the beginning of a query can cause severe perturbation, with F1-Scores exceeding 0.92 across all datasets. (III) Parameter-Scale Vulnerability: Counterintuitively, models with larger parameters exhibit greater vulnerability to the interference. We provide an in-depth analysis to uncover the underlying mechanisms of these phenomena. Furthermore, we raise a critical concern regarding the robustness assumption of current RAG systems, envisioning a threat scenario where an adversary exploits this vulnerability to manipulate the RAG system. We evaluate standard defenses and find them insufficient against EmoRAG. To address this, we propose targeted defenses, analyzing their strengths and limitations in mitigating emoticon-based perturbations. Finally, we outline future directions for building robust RAG systems.

cs.CR↗

arXiv2025

Generalized Divergence Measures and Weak Convergence for the Sets of Probability Measures

This paper extends the asymmetric Kullback-Leibler divergence and symmetric Jensen-Shannon divergence from two probability measures to the case of two sets of probability measures. We establish some fundamental properties of these generalized divergences, including a duality formula and a Pinsker-type inequality. Furthermore, convergence results are derived for both the generalized asymmetric and symmetric divergences, as well as for weak convergence under sublinear expectations.

math.PR↗

arXiv2025

Backdoor Collapse: Eliminating Unknown Threats via Known Backdoor Aggregation in Language Models

Backdoor attacks are a significant threat to large language models (LLMs), often embedded via public checkpoints, yet existing defenses rely on impractical assumptions about trigger settings. To address this challenge, we propose \ourmethod, a defense framework that requires no prior knowledge of trigger settings. \ourmethod is based on the key observation that when deliberately injecting known backdoors into an already-compromised model, both existing unknown and newly injected backdoors aggregate in the representation space. \ourmethod leverages this through a two-stage process: \textbf{first}, aggregating backdoor representations by injecting known triggers, and \textbf{then}, performing recovery fine-tuning to restore benign outputs. Extensive experiments across multiple LLM architectures demonstrate that: (I) \ourmethod reduces the average Attack Success Rate to 4.41\% across multiple benchmarks, outperforming existing baselines by 28.1\%$\sim$69.3\%$\uparrow$. (II) Clean accuracy and utility are preserved within 0.5\% of the original model, ensuring negligible impact on legitimate tasks. (III) The defense generalizes across different types of backdoors, confirming its robustness in practical deployment scenarios.

cs.CL↗

arXiv2025

SaFeR-VLM: Toward Safety-aware Fine-grained Reasoning in Multimodal Models

Multimodal Large Reasoning Models (MLRMs) demonstrate impressive cross-modal reasoning but often amplify safety risks under adversarial or unsafe prompts, a phenomenon we call the \textit{Reasoning Tax}. Existing defenses mainly act at the output level and do not constrain the reasoning process, leaving models exposed to implicit risks. In this paper, we propose SaFeR-VLM, a safety-aligned reinforcement learning framework that embeds safety directly into multimodal reasoning. The framework integrates four components: (I) QI-Safe-10K, a curated dataset emphasizing safety-critical and reasoning-sensitive cases; (II) safety-aware rollout, where unsafe generations undergo reflection and correction instead of being discarded; (III) structured reward modeling with multi-dimensional weighted criteria and explicit penalties for hallucinations and contradictions; and (IV) GRPO optimization, which reinforces both safe and corrected trajectories. This unified design shifts safety from a passive safeguard to an active driver of reasoning, enabling scalable and generalizable safety-aware reasoning. SaFeR-VLM further demonstrates robustness against both explicit and implicit risks, supporting dynamic and interpretable safety decisions beyond surface-level filtering. SaFeR-VLM-3B achieves average performance $70.13$ and $78.97$ on safety and helpfulness across six benchmarks, surpassing both same-scale and $>10\times$ larger models such as Skywork-R1V3-38B, Qwen2.5VL-72B, and GLM4.5V-106B. Remarkably, SaFeR-VLM-7B benefits from its increased scale to surpass GPT-5-mini and Gemini-2.5-Flash by \num{6.47} and \num{16.76} points respectively on safety metrics, achieving this improvement without any degradation in helpfulness performance. Our codes are available at https://github.com/HarveyYi/SaFeR-VLM.

cs.LG↗

arXiv2025

AgentAsk: Multi-Agent Systems Need to Ask

Multi-agent systems (MAS) built on large language models promise improved problem-solving through collaboration, yet they often fail to consistently outperform strong single-agent baselines due to error propagation at inter-agent message handoffs.In this work, we conduct a systematic empirical analysis of such failures and introduce an edge-level error taxonomy that identifies four dominant error types: Data Gap, Signal Corruption, Referential Drift, and Capability Gap, as primary sources of failure in multi-agent interactions. Building on this taxonomy, we propose AgentAsk, a lightweight clarification module designed to intervene at the edge level in MAS to prevent cascading errors. The module operates by strategically applying minimal clarifications at critical points within the system, improving the accuracy and efficiency of the overall task. AgentAsk is trained to balance the trade-offs between clarification cost, latency, and accuracy, while it is also architecture-agnostic and can be easily integrated into existing systems. Evaluated across five benchmarks, AgentAsk consistently improves accuracy by up to 4.69%, while keeping latency and extra costs below 10% compared to baseline MAS, showcasing its high efficiency and minimal overhead.

cs.AI↗

arXiv2025

OrthAlign: Orthogonal Subspace Decomposition for Non-Interfering Multi-Objective Alignment

Large language model (LLM) alignment faces a critical dilemma when addressing multiple human preferences: improvements in one dimension frequently come at the expense of others, creating unavoidable trade-offs between competing objectives like helpfulness and harmlessness. While prior work mainly focuses on constraint-based optimization algorithms and data selection strategies to mitigate conflicts, these approaches overlook the fundamental issue of resolving conflicts directly at the parameter level. In this paper, we present OrthAlign, an innovative approach that pioneers a new paradigm by leveraging orthogonal subspace decomposition to fundamentally resolve gradient-level conflicts in multi-objective preference alignment. OrthAlign strategically decomposes parameter update spaces into orthogonal subspaces, ensuring that optimization toward different preferences occurs in mathematically non-interfering directions. Building upon this, we provide theoretical guarantees demonstrating that when parameter increments satisfy both orthogonal subspace constraints and spectral norm bounds, the resulting updates exhibit linear Lipschitz growth rather than exponential instability, ensuring stable convergence across all preference dimensions. Extensive experiments show that: I. OrthAlign achieves maximum single-preference improvements ranging from 34.61% to 50.89% after multiple-objective alignment across helpful, harmless, and truthful dimensions. II. With an average overall reward improvement of 13.96%.

cs.LG↗

arXiv2025

Backdoor Attribution: Elucidating and Controlling Backdoor in Language Models

Fine-tuned Large Language Models (LLMs) are vulnerable to backdoor attacks through data poisoning, yet the internal mechanisms governing these attacks remain a black box. Previous research on interpretability for LLM safety tends to focus on alignment, jailbreak, and hallucination, but overlooks backdoor mechanisms, making it difficult to understand and fully eliminate the backdoor threat. In this paper, aiming to bridge this gap, we explore the interpretable mechanisms of LLM backdoors through Backdoor Attribution (BkdAttr), a tripartite causal analysis framework. We first introduce the Backdoor Probe that proves the existence of learnable backdoor features encoded within the representations. Building on this insight, we further develop Backdoor Attention Head Attribution (BAHA), efficiently pinpointing the specific attention heads responsible for processing these features. Our primary experiments reveals these heads are relatively sparse; ablating a minimal \textbf{$\sim$ 3%} of total heads is sufficient to reduce the Attack Success Rate (ASR) by \textbf{over 90%}. More importantly, we further employ these findings to construct the Backdoor Vector derived from these attributed heads as a master controller for the backdoor. Through only \textbf{1-point} intervention on \textbf{single} representation, the vector can either boost ASR up to \textbf{$\sim$ 100% ($\uparrow$)} on clean inputs, or completely neutralize backdoor, suppressing ASR down to \textbf{$\sim$ 0% ($\downarrow$)} on triggered inputs. In conclusion, our work pioneers the exploration of mechanistic interpretability in LLM backdoors, demonstrating a powerful method for backdoor control and revealing actionable insights for the community.

cs.CR↗

arXiv2025

Hidden in the Noise: Unveiling Backdoors in Audio LLMs Alignment through Latent Acoustic Pattern Triggers

As Audio Large Language Models (ALLMs) emerge as powerful tools for speech processing, their safety implications demand urgent attention. While considerable research has explored textual and vision safety, audio's distinct characteristics present significant challenges. This paper first investigates: Is ALLM vulnerable to backdoor attacks exploiting acoustic triggers? In response to this issue, we introduce Hidden in the Noise (HIN), a novel backdoor attack framework designed to exploit subtle, audio-specific features. HIN applies acoustic modifications to raw audio waveforms, such as alterations to temporal dynamics and strategic injection of spectrally tailored noise. These changes introduce consistent patterns that an ALLM's acoustic feature encoder captures, embedding robust triggers within the audio stream. To evaluate ALLM robustness against audio-feature-based triggers, we develop the AudioSafe benchmark, assessing nine distinct risk types. Extensive experiments on AudioSafe and three established safety datasets reveal critical vulnerabilities in existing ALLMs: (I) audio features like environment noise and speech rate variations achieve over 90% average attack success rate. (II) ALLMs exhibit significant sensitivity differences across acoustic features, particularly showing minimal response to volume as a trigger, and (III) poisoned sample inclusion causes only marginal loss curve fluctuations, highlighting the attack's stealth.

cs.SD↗