arXiv subjects

Haoyu Liu

Publications and source records attributed to Haoyu Liu.

At least 19 recordsLinked to original sources

arXiv2026

Efficient Test-Time Optimization for Multi-Agent Proof Autoformalization

Full-proof autoformalization bridges extensive mathematical proofs in natural language with formally validated reasoning, offering a pathway to elevate the ceiling of verifiable mathematical reasoning. Unlike statement-level formalization, proof autoformalization is a long-horizon challenge requiring coordination of claims, contexts, and dependencies across many proof steps, yet has only recently come under focused study. Current approaches either rely on costly model training or apply excessive, unguided repair at inference time. To this end, we introduce ToMap, a multi-agent framework that structures proof autoformalization as a Decomposer-Formalizer-Prover pipeline with efficient test-time optimization guided by formal verification and semantic rubrics for proof quality. Rather than distributing test-time compute across all agents, we perform bottleneck analysis and identify the Decomposer as the critical bottleneck: the quality of its atomic, self-contained proof units directly determines whether downstream agents can successfully formalize and prove each step. ToMap therefore treats the Formalizer and Prover as downstream executors and efficiently focuses test-time compute on Decomposer refinement. This refinement follows a loop inspired by GEPA, evolving prompts over candidate decompositions and using formal verification progress together with semantic proof rubrics to define a Pareto frontier that guides the next decomposition update. Experiments on ProofFlowBench show that ToMap improves over the best previous method by 19.0% when evaluated by both syntactic correctness and semantic faithfulness, while requiring lower test-time cost. Scaling analysis shows that most gains emerge within a few iterations of decomposition evolution, guiding test-time budget selection.

cs.AI↗

arXiv2026

Hidden Decoding at Scale: Latent Computation Scaling for Large Language Models

Scaling Large Language Models (LLMs) has been driven mainly by enlarging the Transformer backbone, but for an already-strong model this requires another round of costly pretraining. We study whether an existing backbone can keep improving by allocating more computation to each token while leaving the Transformer backbone fixed. Depth-recurrent (looped) Transformers pursue this goal but are hard to scale, because looped computation does not fit naturally with the pipeline parallelism used to train the largest models. We add computation along the sequence-length dimension, where the extra computation is simply a longer input and stays compatible with standard large-model training. We propose Hidden Decoding, a sequence-length scaling method applied during continued pretraining (CPT). It expands each token into n streams with independent embedding tables and keeps the intermediate streams' key-value cache as context, so each token performs more internal computation without adding or widening Transformer layers. To keep this affordable at scale, we introduce Stream-Factorized Attention, in which most layers attend only within each stream and only a few layers mix across streams, reducing the attention cost from quadratic to roughly linear in n. Experiments support two scaling results. At frontier scale, we train WeLM-HD4-80B and WeLM-HD4-617B at n=4 and improve their matched non-HD baselines, making Hidden Decoding the first demonstrated sequence-length scaling method at the 100B+ MoE scale. Across expansion factors, the gains grow as n increases, showing that sequence-length expansion is a practical fixed-backbone scaling path for frontier-scale LLMs.

cs.CL↗

arXiv2026

Off the Rails: Hijacking the Scoring Head in Generative End-to-End Driving Planners with Safety-Violating Adversarial Perturbations

Generative models have recently seen rapid adoption in End-to-End (E2E) autonomous driving (AD), with diffusion-based denoising and vocabulary-based retrieval becoming the dominant trajectory-decoding paradigms. Despite their architectural diversity, current generative AD planners share a common inference pattern: a fixed set of candidate trajectories (anchors, vocabulary entries, or proposal queries) is scored by one or more learned heads conditioned on the Bird's-Eye-View (BEV) features, and the highest-scored candidate is returned as the final trajectory. Under this design, the scoring head is the only barrier between perception and the motion command, and its decision margins between competing candidates are often small. We introduce \textsc{Derail}, an adversarial framework that exploits this scoring-head attack surface. Evaluated on various generative planners, \textsc{Derail} flips the trajectory selection from a safe to an unsafe candidate, with score drops of $39$--$80\%$ and collision rates of up to $50\%$, consistently outperforming generic loss-maximization and feature-divergence attacks. Our analysis suggests that safety-violating objectives govern attack effectiveness against generative AD planners, and that the scoring-head inference pattern itself is a recurring attack surface worth explicit defensive consideration.

cs.RO↗

arXiv2026

STREAM: A Data-Centric Framework for Mining High-Value Task-Oriented Dialogues from Streaming Media

Large language models for vertical domains are bottlenecked by the scarcity of complex, domain-specific task-oriented dialogues. Existing data acquisition pipelines face a persistent trilemma: expert annotation is expensive, real-world service conversations are constrained by privacy and commercial restrictions, and static corpora quickly become temporally stale. We propose Stream, a data-centric framework that leverages publicly available streaming media (live streams and short videos) to synthesize high-value service dialogues at scale. Stream mines authentic interaction signals from noisy streams and synthesizes conversations by integrating role-grounded persona construction with Conversational Blueprint construction; it further adopts retrieval-augmented generation (RAG) to support knowledge-aware responses. Based on Stream, we release StreamDial, a large-scale multi-domain dataset covering Automotive, Restaurant, and Hotel. StreamDial contains 87,498 dialogue sessions and 1,497,320 turns in total, with an average of 17.11 turns per session and a comparable scale across domains. Each session is organized as a structured quadruplet $\langle P_u, P_a, B, H \rangle$ that pairs dialogue history with explicit user/agent personas and a Conversational Blueprint, capturing realistic service behaviors such as requirement mining, constraint conflicts, negotiation, and recovery. Evaluations with automatic judges and downstream tasks show that StreamDial improves intrinsic dialogue quality over strong baselines, and models trained with StreamDial improve Dialogue State Tracking across backbones; we further report a completed human-evaluation set and encouraging multilingual transfer on Qwen3-8B under a controlled training budget. The data is released in https://github.com/hitxueliang/DialogDataSetBySTREAM.

cs.CL↗

arXiv2026

Imaging Surface Magnetization in Altermagnetic MnTe Films

Altermagnets with pronounced spin-splitting band structure, unconventional magnetic and crystal symmetries, and exotic magneto-transport properties have received immense interest in cutting-edge spintronics, materials science, and condensed matter physics research. Microscopic imaging of spontaneous magnetic domains and phases in altermagnets constitutes an important step for investigating their underlying material properties, mechanisms, and spin behaviors. Taking advantage of scanning-probe quantum microscopy, here we report nanoscale quantum sensing of a prototypical altermagnet candidate $\alpha$-MnTe. We visualize evanescent magnetization and the associated magnetic domains in epitaxial MnTe films, which allows external magnetic fields to control the intrinsic altermagnetic order and configurations. By evaluating a series of MnTe films with different thicknesses down to the atomic scale, we further present evidence for the interfacial origin of the observed weak magnetization and show its correlation with the anomalous Hall effect in MnTe film. Our results advance the current understanding of emergent altermagnetism, providing insights into future material design of altermagnet-integrated spintronic devices.

cond-mat.mtrl-sci↗

arXiv2026

Out of Sight, Out of Track: Adversarial Attacks on Propagation-based Multi-Object Trackers via Query State Manipulation

Recent Tracking-by-Query-Propagation (TBP) methods have advanced Multi-Object Tracking (MOT) by enabling end-to-end (E2E) pipelines with long-range temporal modeling. However, this reliance on query propagation introduces unexplored architectural vulnerabilities to adversarial attacks. We present FADE, a novel attack framework designed to exploit these specific vulnerabilities. FADE employs two attack strategies targeting core TBP mechanisms: (i) Temporal Query Flooding: Generates spurious temporally consistent track queries to exhaust the tracker's limited query budget, forcing it to terminate valid tracks. (ii) Temporal Memory Corruption: Directly attacks the query updater's memory by severing temporal links via state de-correlation and erasing the learned feature identity of matched tracks. Furthermore, we introduce a differentiable pipeline to optimize these attacks for physical-world realizability by leveraging simulations of advanced perception sensor spoofing. Experiments on MOT17 and MOT20 benchmarks demonstrate that FADE is highly effective against state-of-the-art TBP trackers, causing significant identity switches and track terminations.

cs.CV↗

arXiv2026

Dual-Modality Multi-Stage Adversarial Safety Training: Robustifying Multimodal Web Agents Against Cross-Modal Attacks

Multimodal web agents that process both screenshots and accessibility trees are increasingly deployed to interact with web interfaces, yet their dual-stream architecture opens an underexplored attack surface: an adversary who injects content into the webpage DOM simultaneously corrupts both observation channels with a consistent deceptive narrative. Our vulnerability analysis on MiniWob++ reveals that attacks including a visual component far outperform text-only injections, exposing critical gaps in text-centric VLM safety training. Motivated by this finding, we propose Dual-Modality Multi-Stage Adversarial Safety Training (DMAST), a framework that formalizes the agent-attacker interaction as a two-player zero-sum Markov game and co-trains both players through a three-stage pipeline: (1) imitation learning from a strong teacher model, (2) oracle-guided supervised fine-tuning that uses a novel zero-acknowledgment strategy to instill task-focused reasoning under adversarial noise, and (3) adversarial reinforcement learning via Group Relative Policy Optimization (GRPO) self-play. On out-of-distribution tasks, DMAST substantially mitigates adversarial risks while simultaneously doubling task completion efficiency. Our approach significantly outperforms established training-based and prompt-based defenses, demonstrating genuine co-evolutionary progress and robust generalization to complex, unseen environments.

cs.LG↗

arXiv2026

Spiral RoPE: Rotate Your Rotary Positional Embeddings in the 2D Plane

Rotary Position Embedding (RoPE) is the de facto positional encoding in large language models due to its ability to encode relative positions and support length extrapolation. When adapted to vision transformers, the standard axial formulation decomposes two-dimensional spatial positions into horizontal and vertical components, implicitly restricting positional encoding to axis-aligned directions. We identify this directional constraint as a fundamental limitation of the standard axial 2D RoPE, which hinders the modeling of oblique spatial relationships that naturally exist in natural images. To overcome this limitation, we propose Spiral RoPE, a simple yet effective extension that enables multi-directional positional encoding by partitioning embedding channels into multiple groups associated with uniformly distributed directions. Each group is rotated according to the projection of the patch position onto its corresponding direction, allowing spatial relationships to be encoded beyond the horizontal and vertical axes. Across a wide range of vision tasks including classification, segmentation, and generation, Spiral RoPE consistently improves performance. Qualitative analysis of attention maps further show that Spiral RoPE exhibits more concentrated activations on semantically relevant objects and better respects local object boundaries, highlighting the importance of multi-directional positional encoding in vision transformers.

cs.CV↗

arXiv2025

Charged Dirac perturbations on Reissner-Nordstr\"om black holes in a cavity: quasinormal modes with Robin boundary conditions

We investigate charged Dirac quasinormal spectra on Reissner-Nordstr\"om black holes in a mirror-like cavity. For this purpose, we first derive charged Dirac equations, and \textit{two} sets of Robin boundary conditions following the vanishing energy flux principle. The Dirac spectra are then computed both analytically and numerically. Our results reveal a symmetry hidden in the Dirac spectra between two boundary conditions. Moreover, when the cavity is placed close to the event horizon $r_+$, we identify that, in the neutral background the Dirac spectra asymptote to $-(3/8+N/2)i$ [$-(1/8+N/2)i$] for the first [second] boundary condition; while in the charged background the real part of charged Dirac spectra asymptote to $qQ/r_+$ for both boundary conditions; where $N$ is the overtone number, $q$ and $Q$ are charges for the field and for the background. In particular, we uncover a striking anomalous decay pattern, $i.e.$ the excited modes decay \textit{slower} than the fundamental mode, when the charge coupling $qQ$ is large. Our results further illustrate the robustness of vanishing energy flux principle, which are applicable not only to anti-de Sitter black holes but also to black holes in a cavity.

gr-qc↗

arXiv2025

Evolving fractal dimensions in iterative bicolored percolation

Criticality is traditionally regarded as an unstable, fine-tuned fixed point of the renormalization group. We introduce an iterative bicolored percolation process in two dimensions and show that it can both preserve criticality and transform fractal dimensions. Starting from critical configurations, such as the O$(n)$ loop and fuzzy Potts models, successive coarse-graining generates a hierarchy of distinct yet critical generations. Using the conformal loop ensemble, we derive exact, generation-dependent fractal dimensions, which are quantitatively confirmed by large-scale Monte Carlo simulations. The evolutionary trajectory depends not only on the universality class of the initial state but also on whether it possesses a two-state critical structure, leading to different critical exponents starting from site and bond percolation. These results establish a general geometric mechanism for evolving fractal dimensions, in which scale invariance persists across generations.

cond-mat.stat-mech↗

arXiv2025

MME-RAG: Multi-Manager-Expert Retrieval-Augmented Generation for Fine-Grained Entity Recognition in Task-Oriented Dialogues

Fine-grained entity recognition is crucial for reasoning and decision-making in task-oriented dialogues, yet current large language models (LLMs) continue to face challenges in domain adaptation and retrieval controllability. We introduce MME-RAG, a Multi-Manager-Expert Retrieval-Augmented Generation framework that decomposes entity recognition into two coordinated stages: type-level judgment by lightweight managers and span-level extraction by specialized experts. Each expert is supported by a KeyInfo retriever that injects semantically aligned, few-shot exemplars during inference, enabling precise and domain-adaptive extraction without additional training. Experiments on CrossNER, MIT-Movie, MIT-Restaurant, and our newly constructed multi-domain customer-service dataset demonstrate that MME-RAG performs better than recent baselines in most domains. Ablation studies further show that both the hierarchical decomposition and KeyInfo-guided retrieval are key drivers of robustness and cross-domain generalization, establishing MME-RAG as a scalable and interpretable solution for adaptive dialogue understanding.

cs.CL↗

arXiv2025

Who's Who? LLM-assisted Software Traceability with Architecture Entity Recognition

Identifying architecturally relevant entities in textual artifacts is crucial for Traceability Link Recovery (TLR) between Software Architecture Documentation (SAD) and source code. While Software Architecture Models (SAMs) can bridge the semantic gap between these artifacts, their manual creation is time-consuming. LLMs offer new capabilities for extracting architectural entities from SAD and source code to construct SAMs automatically or establish direct trace links. This paper extends our ICSA 2025 paper [19], which introduced Extracting Architecture (ExArch) for LLM-based architecture component name extraction. The extension contributes the novel Architecture Traceability with Entity Matching via Semantic inference (ArTEMiS) approach, an extended evaluation with additional LLMs, configurations, a revised benchmark, and a combined evaluation of both approaches. Specifically, this paper presents the following approaches: ExArch extracts component names as simple SAMs from SAD and source code to eliminate the need for manual SAM creation, while ArTEMiS identifies architectural entities in documentation and matches them with (manually or automatically generated) SAM entities. Our evaluation compares against state-of-the-art approaches SWATTR, TransArC and ArDoCode. TransArC achieves strong performance (F1: 0.87) but requires manually created SAMs; ExArch achieves comparable results (F1: 0.86) using only SAD and code. ArTEMiS is on par with the traditional heuristic-based SWATTR (F1: 0.81) and can successfully replace it when integrated with TransArC. The combination of ArTEMiS and ExArch outperforms ArDoCode, the best baseline without manual SAMs. Our results demonstrate that LLMs can effectively identify architectural entities in textual artifacts, enabling automated SAM generation and TLR, making architecture-code traceability more practical and accessible.

cs.SE↗

arXiv2025

Right Answer at the Right Time - Temporal Retrieval-Augmented Generation via Graph Summarization

Question answering in temporal knowledge graphs requires retrieval that is both time-consistent and efficient. Existing RAG methods are largely semantic and typically neglect explicit temporal constraints, which leads to time-inconsistent answers and inflated token usage. We propose STAR-RAG, a temporal GraphRAG framework that relies on two key ideas: building a time-aligned rule graph and conducting propagation on this graph to narrow the search space and prioritize semantically relevant, time-consistent evidence. This design enforces temporal proximity during retrieval, reduces the candidate set of retrieval results, and lowers token consumption without sacrificing accuracy. Compared with existing temporal RAG approaches, STAR-RAG eliminates the need for heavy model training and fine-tuning, thereby reducing computational cost and significantly simplifying deployment.Extensive experiments on real-world temporal KG datasets show that our method achieves improved answer accuracy while consuming fewer tokens than strong GraphRAG baselines.

cs.IR↗

arXiv2025

Schramm-Loewner evolution contains a topological Sierpiński carpet when $κ$ is close to 8

We consider the Schramm-Loewner evolution (SLE$_κ$) for $κ\in (4,8)$, which is the regime where the curve is self-intersecting but not space-filling. We show that there exists $δ_0>0$ such that for $κ\in (8 - δ_0,8)$, the range of an SLE$_κ$ curve almost surely contains a topological Sierpiński carpet. Combined with a result of Ntalampekos (2021), this implies that in this parameter range, SLE$_κ$ is almost surely conformally non-removable, and the conformal welding problem for SLE$_κ$ does not have a unique solution. Our result also implies that for $κ\in (8 - δ_0,8)$, the adjacency graph of the complementary connected components of the SLE$_κ$ curve is disconnected.

math.PR↗

arXiv2025

Three-point connectivity constant for $q$-state Potts spin clusters

Recently, Ang--Cai--Sun--Wu (2024) determined the three-point connectivity constant for two-dimensional critical percolation, confirming a prediction of Delfino and Viti (2010). In this paper, we address the analogous problem for planar critical $q$-state Potts spin clusters. We introduce a continuum three-point connectivity constant and compute it explicitly. Under the scaling-limit conjecture for Potts spin clusters, this quantity coincides with the scaling limit of the properly normalized probability that three points lie in the same spin cluster. The resulting formula agrees with the imaginary DOZZ formula up to an explicit $q$-dependent constant with a geometric interpretation. This answers a question from Delfino--Picco--Santachiara--Viti (2013). The proof exploits the coupling between CLE and LQG, together with the BCLE descriptions of $q$-state Potts scaling limits due to Miller--Sheffield--Werner (2017) and Köhler-Schindler and Lehmkühler (2025).

math.PR↗

arXiv2025

See No Evil: Adversarial Attacks Against Linguistic-Visual Association in Referring Multi-Object Tracking Systems

Language-vision understanding has driven the development of advanced perception systems, most notably the emerging paradigm of Referring Multi-Object Tracking (RMOT). By leveraging natural-language queries, RMOT systems can selectively track objects that satisfy a given semantic description, guided through Transformer-based spatial-temporal reasoning modules. End-to-End (E2E) RMOT models further unify feature extraction, temporal memory, and spatial reasoning within a Transformer backbone, enabling long-range spatial-temporal modeling over fused textual-visual representations. Despite these advances, the reliability and robustness of RMOT remain underexplored. In this paper, we examine the security implications of RMOT systems from a design-logic perspective, identifying adversarial vulnerabilities that compromise both the linguistic-visual referring and track-object matching components. Additionally, we uncover a novel vulnerability in advanced RMOT models employing FIFO-based memory, whereby targeted and consistent attacks on their spatial-temporal reasoning introduce errors that persist within the history buffer over multiple subsequent frames. We present VEIL, a novel adversarial framework designed to disrupt the unified referring-matching mechanisms of RMOT models. We show that carefully crafted digital and physical perturbations can corrupt the tracking logic reliability, inducing track ID switches and terminations. We conduct comprehensive evaluations using the Refer-KITTI dataset to validate the effectiveness of VEIL and demonstrate the urgent need for security-aware RMOT designs for critical large-scale applications.

cs.CV↗

arXiv2025

A Comprehensive Benchmark on Spectral GNNs: The Impact on Efficiency, Memory, and Effectiveness

With recent advancements in graph neural networks (GNNs), spectral GNNs have received increasing popularity by virtue of their ability to retrieve graph signals in the spectral domain. These models feature uniqueness in efficient computation as well as rich expressiveness, which stems from advanced management and profound understanding of graph data. However, few systematic studies have been conducted to assess spectral GNNs, particularly in benchmarking their efficiency, memory consumption, and effectiveness in a unified and fair manner. There is also a pressing need to select spectral models suitable for learning specific graph data and deploying them to massive web-scale graphs, which is currently constrained by the varied model designs and training settings. In this work, we extensively benchmark spectral GNNs with a focus on the spectral perspective, demystifying them as spectral graph filters. We analyze and categorize 35 GNNs with 27 corresponding filters, spanning diverse formulations and utilizations of the graph data. Then, we implement the filters within a unified spectral-oriented framework with dedicated graph computations and efficient training schemes. In particular, our implementation enables the deployment of spectral GNNs over million-scale graphs and various tasks with comparable performance and less overhead. Thorough experiments are conducted on the graph filters with comprehensive metrics on effectiveness and efficiency, offering novel observations and practical guidelines that are only available from our evaluations across graph scales. Different from the prevailing belief, our benchmark reveals an intricate landscape regarding the effectiveness and efficiency of spectral graph filters, demonstrating the potential to achieve desirable performance through tailored spectral manipulation of graph data.

cs.LG↗

arXiv2025

When Deepfake Detection Meets Graph Neural Network:a Unified and Lightweight Learning Framework

The proliferation of generative video models has made detecting AI-generated and manipulated videos an urgent challenge. Existing detection approaches often fail to generalize across diverse manipulation types due to their reliance on isolated spatial, temporal, or spectral information, and typically require large models to perform well. This paper introduces SSTGNN, a lightweight Spatial-Spectral-Temporal Graph Neural Network framework that represents videos as structured graphs, enabling joint reasoning over spatial inconsistencies, temporal artifacts, and spectral distortions. SSTGNN incorporates learnable spectral filters and spatial-temporal differential modeling into a unified graph-based architecture, capturing subtle manipulation traces more effectively. Extensive experiments on diverse benchmark datasets demonstrate that SSTGNN not only achieves superior performance in both in-domain and cross-domain settings, but also offers strong efficiency and resource allocation. Remarkably, SSTGNN accomplishes these results with up to 42$\times$ fewer parameters than state-of-the-art models, making it highly lightweight and resource-friendly for real-world deployment.

cs.CV↗