Search arXiv⌕ Search

arXiv subjects

Jun Luo

Publications and source records attributed to Jun Luo.

At least 19 recordsLinked to original sources
arXiv2026

LangMap: A Human-Verified Benchmark for Hierarchical Open-Vocabulary Goal Navigation

Language-conditioned goal navigation requires embodied agents to locate user-specified targets without step-by-step guidance. However, existing benchmarks largely focus on category-level goals or rely on instance descriptions generated by vision-language models, which often contain ambiguities and semantic errors, limiting systematic and reliable evaluation. We introduce HieraNav, an open-vocabulary navigation task with goals at four hierarchical semantic levels: scene, room, region, and instance. We present Language as a Map (LangMap), the first goal navigation benchmark to enrich real-world indoor 3D scans with human-verified semantic annotations supporting tasks across all four goal levels. Built on HM3D using a contrastive annotation protocol that compares same-scene regions and instances, LangMap provides region labels and discriminative region and instance descriptions covering 414 object categories and contains over 18K tasks. Each target has concise and detailed descriptions, enabling evaluation across instruction styles. Automated and human evaluations validate our annotation quality: our descriptions improve text-to-view matching accuracy over GOAT-Bench's by 23 points on all shared annotated instances, and an independent human audit yields 92.5% unique-and-correct matches. We also propose PlaNaVid, an RGB-only baseline that combines Bounded Diverse Memory with high-level planning to prime a reactive policy for multi-goal navigation, achieving top-tier success rates without depth, 3D scene representations, or object masks. Further analyses reveal that exploration and hierarchical disambiguation failures become more prominent at finer goal levels, while long-tail categories, small objects, distant targets, timely stopping, and multi-goal completion remain challenging. Benchmark and code: https://bo-miao.github.io/LangMap

cs.CV↗
arXiv2026

TIC-GRPO: Provable and Efficient Optimization for Reinforcement Learning from Human Feedback

Group Relative Policy Optimization (GRPO) trains language models without a value critic using rewards centered within sampled response groups. We study how importance weighting, clipping, and length normalization affect its stochastic updates and propose Trajectory-level Importance-Corrected GRPO (TIC-GRPO), combining an upper importance-ratio cap with a full-trajectory likelihood ratio. Under common maximum-length normalization, a trajectory change of measure gives a score second moment proportional to the response horizon. We propagate this estimate through the actual empirical baseline and all inner updates. With separately optimized constant steps, the sampling term in TIC's stationarity bound has a linear horizon coefficient, versus order three halves for token-level GRPO$_2$. Matching upper and lower bounds separate the capped methods' worst-case response-level variances by a factor $T$. The bounds track prompt count, response-group size, vocabulary size, reward scale, and score regularity; per-response-normalized GRPO also retains a length-covariance term. Experiments at two Qwen3 scales on four reasoning and coding benchmarks evaluate both mechanisms in sampled-token-normalized training.

cs.LG↗
arXiv2026

RetryRoute: Retry-Aware Model Routing for Agentic Image Editing Services

Web agents increasingly edit images by calling open editing models in a critique-and-retry loop, and these models differ in cost by more than an order of magnitude. After every rejected edit, a service must decide whether to retry the same model, escalate to another or give up. Existing routers pick one model per request, and cascades escalate in a fixed order. However, both treat a retry as an independent draw. We identify sticky failures, meaning that a retry after a failure succeeds only 17-53% as often as a first try across six open models, three benchmarks and two judges. Based on this finding, we propose RetryRoute, a retry-aware router that estimates each model's chance of success given the request's failure history. It plans retries, escalations and abandonment over failure states and adds live queue waits to the cost, so the same plan becomes load-aware dispatch. At the main budget, RetryRoute's success under the in-loop judge exceeds the strongest of eleven prior baselines by 2.8 points on average over eight groups. On a GPU serving testbed it raises the share of edits delivered within the service-level objective by 5.0-6.3 points over the best prior baseline at every load.

cs.LG↗
arXiv2026

When Links Mislead: Label-Free Structural Polarity for Zero-Shot Classification of Web Pages

Web platforms such as digital libraries, online stores and university sites must classify new pages before anyone labels them. Large language models (LLMs) can label each page from its text, but these labels are noisy, and links could correct them only where linked pages share classes. Existing zero-shot methods ignore the links, or pull a page toward its neighbours as label propagation and graph neural networks (GNNs) do, which hurts where linked pages differ in class. Others weight propagation by an LLM-estimated homophily that costs extra calls and drifts across LLMs. We find that the sign with which links should enter a prediction, which we call structural polarity, is readable without true labels. It is the excess agreement of LLM labels and text-prototype predictions along links. We thus propose ZeroPolar, which gates one signed propagation step by this statistic. Training-free ZeroPolar mixes prototype scores with any number of LLM labels, and ZeroPolar-T denoises the labels with a text classifier once half of the pages carry one. With one GPT-4o label per page, ZeroPolar-T reaches the highest mean accuracy on eleven graphs (73.99%), 6.01 points above LLM-BP and significantly above all eighteen baselines we compare. With LLM labels on only 5% of the pages, training-free ZeroPolar leads a GNN trained on these labels by 10.80 points.

cs.LG↗
arXiv2026

Redundancy Meets Synergy: Dependency-aware Expert Selection for MoE via Submodular Optimization

While Mixture-of-Experts (MoE) models effectively scale model capacity through sparse activation, their deployment is often bottlenecked by prohibitive memory requirements. Extracting a compact subset of experts presents a promising solution. However, existing expert selection heuristics predominantly rely on Top-$k$ ranking, which isolates the evaluation of individual experts and ignores the intricate inter-expert dependencies introduced by the MoE gating network. In this paper, we propose DS-MoE, a theoretically grounded framework that redefines expert selection via difference-of-submodular (DS) optimization. By analyzing the second-order Taylor expansion of the loss degradation, we reveal functional duality within expert combinations: redundancy (where experts encode overlapping representations) and synergy (where experts provide complementary error cancellation). To navigate this duality, we mathematically decouple redundancy reduction from synergy maximization by formulating the selection objective as a DS function. Furthermore, we devise a tailored majorization-minimization (MM) algorithm with provable guarantees to efficiently identify a high-quality expert subset. Extensive experiments demonstrate that DS-MoE effectively preserves indispensable expert combinations, achieving superior performance compared to the state-of-the-art baselines.

cs.LG↗
arXiv2026

Physically-Induced Atmospheric Adversarial Perturbations: Enhancing Transferability and Robustness in Remote Sensing Image Classification

Adversarial attacks pose a severe threat to the reliability of deep learning models in remote sensing (RS) image classification. Most existing methods rely on direct pixel-wise perturbations, failing to exploit the inherent atmospheric characteristics of RS imagery or survive real-world image degradations. In this paper, we propose FogFool, a physically plausible adversarial framework that generates fog-based perturbations by iteratively optimizing atmospheric patterns based on Perlin noise. By modeling fog formations with natural, irregular structures, FogFool generates adversarial examples that are not only visually consistent with authentic RS scenes but also deceptive. By leveraging the spatial coherence and mid-to-low-frequency nature of atmospheric phenomena, FogFool embeds adversarial information into structural features shared across diverse architectures. Extensive experiments on two benchmark RS datasets demonstrate that FogFool achieves superior performance: not only does it exceed in white-box settings, but also exhibits exceptional black-box transferability (reaching 83.74% TASR) and robustness against common preprocessing-based defenses such as JPEG compression and filtering. Detailed analyses, including confusion matrices and Class Activation Map (CAM) visualizations, reveal that our atmospheric-driven perturbations induce a universal shift in model attention. These results indicate that FogFool represents a practical, stealthy, and highly persistent threat to RS classification systems, providing a robust benchmark for evaluating model reliability in complex environments.

cs.CV↗
arXiv2026

PACT: Privileged Trace Co-Training for Multi-Turn Tool-Use Agents

Multi-turn tool-use agents must reason, call tools, and adapt to observations across several interaction turns. Post-training such agents is challenging, as reinforcement learning often suffers from sparse rewards and weak credit assignment despite matching the prompt-only inference setting, while supervised fine-tuning on expert traces provides dense process supervision but can over-constrain the model to fixed trajectories. To tackle this, we propose PACT, a Privileged trAce Co-Training framework for multi-turn tool-use agents. The key idea is to use expert traces only as training-time optimization signals rather than rollout-time hints. PACT keeps rollout generation prompt-only, then uses expert traces to guide optimization through two complementary signals: a trace-conditioned RL surrogate that evaluates prompt-only rollouts under expert-trace context, and a component-aware SFT loss that provides annealed process supervision over reasoning and tool-call components. To reduce over-reliance on the training-only trace context, PACT incorporates prompt-only anchoring with standard prompt-only RL updates. Experiments on FTRL, BFCL, and ToolHop show that PACT consistently outperforms strong SFT- and RL-based baselines. Further analyses support the complementarity of the two training objectives and the benefits of component-aware supervision and prompt-only anchoring. These results highlight the effectiveness of privileged trace co-training for multi-turn tool-use learning. The project page is available at https://zhenbangdu.github.io/pact-project-page.

cs.CL↗
arXiv2026

More Than Meets the Eye? Uncovering the Reasoning-Planning Disconnect in Training Vision-Language Driving Models

Vision-Language Model (VLM) driving agents promise explainable end-to-end autonomy by first producing natural-language reasoning and then predicting trajectory planning. However, whether planning is causally driven by this reasoning remains a critical but unverified assumption. To investigate this, we build DriveMind, a large-scale driving Visual Question Answering corpus with plan-aligned Chain-of-Thought (CoT), automatically generated from nuPlan. Our data generation process converts sensors and annotations into structured inputs and, crucially, separates priors from to-be-reasoned signals, enabling clean information ablations. Using DriveMind, we train representative VLM agents with Supervised Fine-Tuning and Group Relative Policy Optimization and evaluate them with nuPlan's metrics. Our results, unfortunately, indicate a consistent causal disconnect in reasoning-planning: removing ego/navigation priors causes large drops in planning scores, whereas removing CoT produces only minor changes. Attention analysis further shows that planning primarily focuses on priors rather than the CoT. Based on this evidence, we propose the Reasoning-Planning Decoupling Hypothesis, positing that the training-yielded reasoning is an ancillary byproduct rather than a causal mediator. To enable efficient diagnosis, we introduce a novel, training-free probe that measures an agent's reliance on priors by evaluating its planning robustness against minor input perturbations. In summary, we provide the community with a new dataset and a diagnostic tool to evaluate the causal fidelity of future models.

cs.AI↗
arXiv2026

Adam under Generalized Smoothness with Second-Moment-Type Stochastic Gradients

Adam is widely observed to remain stable even when the objective deviates significantly from global smoothness. Under the generalized smoothness framework, however, existing analyses rely on strong tail assumptions on the stochastic gradients, such as almost-sure boundedness or sub-Gaussianity. Whether Adam converges on generalized smooth objectives under only second moment information on the stochastic gradients, without such concentration assumptions, was identified as an important open direction by Li et al. (2023). This paper gives an affirmative answer under fairly general conditions: such tail assumptions are not necessary. Building on the Adam self-normalization framework of Jin et al. (2026), developed for classical smoothness and bounded variance, we extend the stopping-time and de-preconditioning strategy to the $L_0$-$L_p$ generalized smoothness condition and a generalized second moment ABC condition. Even when the stochastic-gradient condition provides only second moment information that may grow along the trajectory, the stochastic trajectory of Adam remains in a locally well-behaved smoothness region, with stretched-exponential tail decay under bounded variance and global smoothness. Consequently, we establish high-probability convergence rate guarantees over the full range $p<2$, with confidence dependence of order $δ^{-1/2}$, while the stepsize prefactor depends on $δ$ only through a single logarithmic factor. We further construct a hard instance showing that, under only second-moment information, this $δ^{-1/2}$-type confidence dependence is sharp. Finally, in the regime $p<1$, we combine the trajectory control with polynomial-growth estimates on rare events to obtain convergence rate guarantees in expectation.

cs.AI↗
arXiv2026

Asynchronous LLM Post-Training: Group-Mass Capping and Convergence Analysis

Asynchronous reinforcement learning (RL) improves the efficiency of large language model post-training but introduces stale rollouts generated by earlier policies. Theoretical understanding of how this staleness affects convergence and how to mitigate its impact remains limited. We derive a convergence bound for GRPO-style algorithms that explicitly characterizes the tradeoff between the gradient estimator's second moment and bias. For trajectory-level importance-weighted estimators, our analysis shows that once the second moment is uniformly controlled, delay enters the bound through the bias introduced by clipping or rescaling. Guided by this insight, we propose a novel group mass capping GRPO (GMC-GRPO) method, which minimizes a ratio-based bias bound within a class of weighted estimators sharing a common second-moment guarantee. We establish convergence guarantees for asynchronous GMC-GRPO and show that, compared with TIC-GRPO, it improves the threshold dependence of the fourth-order delay term from $O(ε^{-4})$ to $O(ε^{-2})$ as $ε\to0$, where $1+ε$ is the ratio threshold. Under local policy overlap, the delay-dependent term decreases as $G^{-2/5}$ after tuning the step size, where $G$ is the group size. For fixed behavior and current policies, the bias introduced by group rescaling also vanishes as $G\to\infty$, whereas the bias from trajectory-wise clipping can persist. Experiments across Qwen3 models and reasoning benchmarks demonstrate improved robustness to stale rollouts, with GMC-GRPO achieving the best performance among stable baselines under large rollout delays.

cs.LG↗
arXiv2026

HappyWorld-Bench

Evaluating world models requires assessing both the quality of the worlds they generate and their consistency and responsiveness under exploration, interaction, and modification. We introduce HappyWorld-Bench, a comprehensive benchmark that evaluates whether generated worlds remain reliable as agents interact with them. Our design is built on a hierarchical capability framework of six world capabilities (W1-W6), from generative construction to unified world modeling, instantiated across three independent evaluation tracks: video world models, spatial world models, and embodied world models. HappyWorld-Bench comprises 1,138 video prompts, 300 spatial scenes, and 254 embodied test cases. Across all three tracks, we build and operate HappyWorld-Arena to organize human A/B comparisons and derive model-level Elo ratings, which complement newly designed automated metrics that capture behavioral correctness. We evaluate 14 video world models, 9 spatial systems, and 8 embodied candidates under this unified framework. Results reveal remaining reliability gaps across all three tracks: video models exhibit reduced consistency during extended rollouts and revisits, spatial models achieve at best 70.14% placement accuracy and 73.33% edit execution, and embodied models struggle to preserve state across multi-step actions and respond precisely to altered action conditions and physical rules. These findings highlight the need to evaluate world models not only by visual quality, but also by state consistency and the correctness of their responses to actions and interventions.

cs.CV↗
arXiv2026

Feasibility Distance Fields for Heterogeneous Constraints in Robot Configuration Space

Robot manipulators are monitored by constraint-specific indicators whose units and gradient scales are not comparable, so they do not provide a common measure of the configuration-space motion remaining before violation. We define the feasibility distance field (FDF) as the distance, under a fixed positive-definite joint-space metric, to the union of infeasible configuration sets. Classical distance-to-set theory gives 1-Lipschitz continuity, almost-everywhere differentiability, and unit dual-gradient norm wherever the nearest projection is unique. The robotics contribution is an admissibility analysis showing when practical constraints define non-empty closed sets. We derive admissible formulations for external and self-collision, joint limits, dexterity, Cartesian and task-projected compliance, joint torque under payload, and dynamic manipulability. Since every field uses the same metric, heterogeneous constraints compose by a pointwise minimum, conditioned constraints retain a fixed distance space, and multi-robot constraints produce block-sparse gradients that identify which robots must react. We generate projection-based labels and train neural approximations with a distance loss and an Eikonal penalty. Simulations on a UR5e and a dual-arm cell evaluate seven fields using value, projection, sign, gradient, composition, and moving-obstacle diagnostics. Across 8,000 configurations, the largest feasible-side secant ratio is 0.920, mean learned gradient norms range from 0.994 to 0.998, and projection residuals range from 0.011 to 0.034 rad. Across 24 random obstacle paths, the external and composed collision fields achieve 90.4% and 91.6% success within 3 cm, with sign-error rates below 2%. The results support a common configuration-space margin and identify approximation errors near medial axes and sparsely sampled boundaries.

cs.RO↗
arXiv2026

Secrets in Radio Waves: Towards Practical and Protocol-Agnostic PHY Information Hiding

Physical layer (PHY) information hiding supports critical applications, such as digital fingerprinting for transmitter identification and undetectable side channels for covert communication, and has attracted considerable attention from the research community. One category of prior studies focuses on theoretical analysis, proposing techniques such as artificial noise or reconfigurable intelligent surfaces to enable undetectable covert transmission. However, hardware prototypes are rarely presented due to their algorithmic complexity or hard-to-satisfy assumptions. Another category of studies focuses on system-level solutions, achieving PHY information hiding by customizing existing modulation schemes. However, these methods are typically designed for specific wireless protocols, limiting their generalizability. In this work, we introduce a new PHY information hiding paradigm that differs fundamentally from the previous two categories of approaches. Inspired by recent advancements in other domains such as image information hiding, we migrate encoder-decoder neural networks to the PHY information hiding field, embedding secrets by introducing imperceptible distortions within the preamble waveform. Sim-to-real fine-tuning is additionally proposed to tackle unique challenges, e.g., fading and hardware imperfections. The designed methodology is practical and protocol-agnostic. We provide case hardware prototypes of two commercially popular wireless technologies, i.e., LoRa and Bluetooth Low Energy (BLE), using commodity software-defined radio (SDR) transceivers, demonstrating excellent feasibility and generalizability.

eess.SP↗
arXiv2026

TIGA: Trajectory-Injected Generative Attack against Black-box AIGC Detectors

Recent diffusion models have achieved remarkable realism in facial image synthesis, posing growing challenges to artificial intelligence-generated content (AIGC) forensic detectors.Existing evasion methods typically perturb pre-generated images or require detector-aware training, which may introduce visible or statistical artifacts and limit applicability when the diffusion model must remain frozen and the target detector is accessible only through black-box queries. We propose Trajectory-Injected Generative Attack (TIGA), a source-image-free and training free framework that generates detector-evasive images within a single diffusion sampling trajectory. TIGA steers the latent Denoising Diffusion Implicit Model (DDIM) trajectory so that adversarial properties emerge during generation rather than being added afterward. TIGA first aggregates gradients from multiple white-box surrogate detectors to form a transferable, sign-aware prior, and then performs anisotropic directional search with symmetric finite-difference queries to estimate the black-box target response. The estimated directions are stabilized by decayed momentum and injected according to the DDIM noise schedule, with frequency-domain reshaping to suppress high frequency artifacts. Experiments on surrogate and unseen specialized forensic detectors show that TIGA achieves strong blackbox attack performance, transferability, and high robustness under common post-processing operations without source images or diffusion-model retraining, while preserving high perceptual quality.

cs.CV↗
arXiv2026

Visual Attention Faithfulness in Vision-Language Models is Heterogeneous

Whether attention weights faithfully reflect model reasoning has been actively debated in NLP, yet this question remains largely unexplored for the visual modality in Vision-Language Models (VLMs). We address this gap through causal perturbation analysis on current VLMs, evaluating both the comprehensiveness and sufficiency gap of attention-ranked visual tokens. Our analysis reveals that visual attention faithfulness is heterogeneous, manifesting in three distinct processing modes: Faithful-Sufficient, where top-$k$ attention tokens are both necessary and sufficient for prediction; Faithful-Distributed, where they are necessary but broader visual context remains required; and Non-Focal, where no localized attention region is individually necessary while visual information remains an essential trigger for prediction. Furthermore, human-annotated ground-truth regions satisfy comprehensiveness in only $\sim 60$% of cases compared with model attention rankings, revealing systematic divergence between model visual reliance and human intuition. We demonstrate these patterns across both general VQA on VQAv2 and document tasks on VRDU and ChartQA, showing that visual attention faithfulness varies systematically with processing demands and model architectures rather than being uniformly faithful or unfaithful.

cs.CV↗
arXiv2026

Beyond Token-Level Guidance: Inference-Time Alignment of Specialized LLMs via Cross-Family Representation Steering

Large language models (LLMs) finetuned for specialized domains represent crucial high-impact applications. Inference-time alignment improves safety degraded from specialization finetuning without requiring substantial computational resources, complementing finetuning-based methods with an easy-to-use, plug-and-play solution. However, existing inference-time methods fail to reliably improve safety without disrupting domain capability. We identify the root cause as complementary expertise orthogonality: specialized base models and general-domain guidance models have orthogonal competencies, making the guidance signal unreliable for specialized generation. This primarily manifests as stop token interference, where the guidance model's tendency toward continuation overrides the base model's decision to stop, burying correct answers under guidance-induced continuation. To address this problem, we propose CREST, an inference-time alignment method that steers base model hidden representations using safety directions extracted from a guidance model of any family, avoiding token-level structural limitations entirely. CREST improves safety where specialization has weakened it while preserving both domain-specific capability and the safety of already well-aligned models, outperforming baselines by up to 22.2\% on safety benchmarks. Our code is available at: https://github.com/DecayingSeart/CREST.

cs.CL↗
arXiv2026

Invisible Walls: Privacy-Preserving ISAC Empowered by Reconfigurable Intelligent Surfaces

The environmental and target-related information inherently carried in wireless signals, such as channel state information (CSI), has brought increasing attention to integrated sensing and communication (ISAC). However, it also raises pressing concerns about privacy leakage through eavesdropping. While existing efforts have attempted to mitigate this issue, they either fail to account for the needs of legitimate communication and sensing or rely on hardware with high cost. To overcome these limitations, we propose PrivISAC, a plug-and-play, low-cost solution that leverages reconfigurable intelligent surface (RIS) to protect user privacy while preserving ISAC performance. At its core, PrivISAC constructs a small set of RIS configurations from two beamforming vectors per RIS row and randomly activates one configuration per time slot to perturb CSI and mask sensitive sensing information from eavesdroppers. The two vectors are designed to maintain similar communication-direction responses while creating distinct sensing-direction responses, thereby preserving transmission quality and masking sensitive CSI from eavesdroppers. To enable legitimate sensing, we introduce a time-domain masking and demasking method that maps CSI samples to their RIS configurations and removes configuration-induced discrepancies. We implement PrivISAC on commodity wireless devices and conduct extensive experiments. Results show that PrivISAC provides effective privacy protection while preserving high-quality communication and sensing performance for legitimate receivers.

eess.SP↗
arXiv2026

DeepStream: Prototyping Deep Joint Source-Channel Coding for Real-Time Multimedia Transmissions

Deep learning-based joint source-channel coding (DeepJSCC) has emerged as a promising technique in 6G for enhancing the efficiency and reliability of data transmission across diverse modalities, particularly in low signal-to-noise ratio (SNR) environments. This advantage is realized by leveraging powerful neural networks to learn an optimal end-to-end mapping from the source data directly to the transmit symbol sequence, eliminating the need for separate source coding, channel coding, and modulation. Although numerous efforts have been made towards efficient DeepJSCC, they have largely stayed at numerical simulations that can be far from practice, leaving the real-world viability of DeepJSCC largely unverified. To this end, we prototype DeepStream upon orthogonal frequency division multiplexing (OFDM) technology to offer efficient and robust DeepJSCC for multimedia transmission. In conforming to OFDM, we develop both a feature-to-symbol mapping method and a cross-subcarrier precoding method to improve the subcarrier independence and reduce peak-to-average power ratio. To reduce system complexity and enable flexibility in accommodating varying quality of service requirements, we further propose a progressive coding strategy that adjusts the compression ratio based on latency with minimal performance loss. We implement DeepStream for real-time image transmission and video streaming using software-defined radio. Extensive evaluations verify that DeepStream outperforms both the standard scheme and the direct deployment scheme. Particularly, at an SNR of 10 dB, DeepStream achieves a PSNR of 35 dB for image transmission and an MS-SSIM of 20 dB for video streaming, whereas the standard scheme fails to recover meaningful information.

eess.SP↗