Searcharxiv⌕ Search

arXiv subjects

Xuesong Wang

Publications and source records attributed to Xuesong Wang.

At least 19 recordsLinked to original sources

Marginal-Preserving Coupled Diffusion Composition for Long-Horizon Planning

Compositional diffusion planners extend short-horizon trajectory models to long-horizon goal-conditioned planning by generating and stitching multiple local trajectory chunks. However, in multimodal environments, local plausibility is not sufficient for global consistency: independently initialized chunks may denoise toward incompatible overlap modes, producing plans that are individually feasible but incoherent after stitching. Existing remedies typically add search, guidance, pruning, or graph-based structure at inference time, increasing computational cost or relying on task-specific machinery. We instead investigate graph-structured dependence among chunk-wise initial diffusion noises as a lightweight coordination mechanism for compositional trajectory planning. We propose \textbf{M}arginal-Preserving \textbf{C}oupled \textbf{D}iffusion \textbf{C}omposition (MCDC), a training-free inference method for pretrained compositional diffusion planners. Instead of sampling chunk noises independently, MCDC constructs coupled Gaussian initializations whose per-chunk marginals remain exactly standard Gaussian, as required by the pretrained denoiser, while neighboring chunks receive positively correlated noise components. This preserves the standard-Gaussian initialization marginal of each chunk while biasing adjacent chunks toward compatible overlap-mode choices. We prove exact marginal preservation and show, under a simple multimodal overlap model, that positive coupling increases the probability of compatible mode selection. On OGBench long-horizon stitch tasks, MCDC consistently improves over independent compositional diffusion sampling while retaining nearly the same inference time, and achieves a favorable success--cost trade-off compared with search-based composition methods.

cs.AI↗

Sledgehammer or Scalpel? A Fine-grained Adaptive Framework for Implicit Hate Speech Detection

Unlike explicit attacks with obvious profanity, implicit hate speech hides malice within seemingly compliant expressions through metaphors and contextual hints, making its detection in online content review challenging. Many PLM- or LLM-based methods apply the same processing pipeline across samples, which may overlook differences in information requirements and introduce unnecessary computation. We observe that online hate speech is not monolithic but manifests in varied forms. We therefore define three fine-grained categories: Shallow, Targeted, and Context-Dependent. Accordingly, we propose Fine-grained Adaptive Implicit Hate speech Detection (FAID), a novel framework that first performs fine-grained classification and then adapts to specific categories. Specifically, for Shallow samples with surface-identifiable intents, the framework adopts lightweight Prompt-Tuning for rapid classification; for Targeted comments that bind malicious intent to concealed targets, we design knowledge augmentation to iteratively refine the model and reveal hidden targets; for Context-Dependent comments lacking background information, we utilize an agentic framework that automatically generates prompts to evolve context, infer missing background information and identify ambiguous malicious intents. This adaptive architecture focuses computational resources on complex implicit samples while avoiding redundant reasoning for shallow samples. Experiments on four benchmark datasets demonstrate that FAID significantly outperforms SOTA baselines.

cs.CL↗

Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning

Offline goal-conditioned reinforcement learning (GCRL) learns goal-directed policies from reward-free data, but in long-horizon tasks, goal-conditioned value functions often provide unstable guidance due to sparse rewards and discounting. Hierarchical methods partially mitigate this issue via subgoal decomposition; however, high-level decision-making still relies on noise-sensitive value estimates, leading to unstable behavior in complex environments. We address this limitation by proposing \textbf{D}iffusion \textbf{S}ubgoal \textbf{P}lanning (\textbf{DSP}), a diffusion-based framework for high-level subgoal generation. DSP casts high-level planning as guided generative inference over goal-conditioned subgoals and learns both conditional and unconditional flows, enabling classifier-free guidance to introduce a goal-directed bias at inference time. By removing explicit value-based guidance from high-level planning, DSP generates reachable and goal-directed subgoals through a generative model while retaining hierarchical execution. Experiments on offline GCRL benchmarks demonstrate that DSP outperforms prior methods on a range of navigation and manipulation tasks, with particularly strong performance in maze environments that require multi-step subgoal planning.

cs.AI↗

Diffusion-aided Task-oriented Semantic Communications with Model Inversion Attack

Semantic communication enhances transmission efficiency by conveying semantic information rather than raw input symbol sequences. Task-oriented semantic communication further aims to retain only task-specific information, thereby achieving greater bandwidth savings. However, these neural-network-based communication systems are vulnerable to model inversion attacks, in which adversaries attempt to recover sensitive input information from intercepted semantic features. The key challenge is therefore to preserve privacy while maintaining task accuracy and robustness. We consider a task-confidential setting in which the adversary attempts to reconstruct the original input from intercepted features without knowing the legitimate receiver's task or model. Although PSNR and SSIM are commonly used to assess reconstruction quality, we find that an external classifier can still perform the legitimate receiver's task with nontrivial accuracy on reconstructions with low PSNR or SSIM, indicating that these reconstructions still contain task-level semantic leakage. We therefore propose DiffSem, which splits the diffusion process between controlled transmitter-side self-noising and matched receiver-side reverse denoising. Experiments on the MNIST, CIFAR-10, and CelebA datasets show that DiffSem improves the legitimate receiver's task accuracy without increasing either the transmitted feature size or information leakage.

cs.CR↗

Risk-Averse Power System Resilience Planning Under AI Data Center Demand Growth Using a Two-Stage DRO-CVaR MILP Optimization

The rapid growth of artificial intelligence-data centers is introducing significant demand variability and operational uncertainty into modern power systems, creating new challenges for resilience planning and grid operation. In addition to physical disruptions such as line outages and uncertain restoration processes, AI-driven loads can exhibit highly dynamic behavior. This paper presents a two-stage risk-averse Distributionally Robust Optimization (DRO)-Mixed Integer Linear Programming framework for enhancing distribution-system resilience under combined AI-driven demand uncertainty and physical disruptions. The proposed framework coordinates the strategic prepositioning and adaptive dispatch of flexible capacity modules, including battery energy storage systems and mobile diesel generation resources. A unified scenario representation captures AI-driven demand variability, line outages, and repair-time uncertainty. To improve robustness against uncertainty misspecification and extreme events, the model integrates Conditional Value-at-Risk within a DRO formulation. Validation on the IEEE 33-bus distribution system demonstrates reductions in Energy Not Supplied, improved served-load recovery, enhanced restoration performance, and increased system resilience under severe operating conditions.

eess.SY↗

SAM3 Self-Distillation for Fine-Grained GOOSE 2D Semantic Segmentation

We describe our 4th-place entry to the ICRA 2026 GOOSE 2D Fine-Grained Semantic Segmentation Challenge, which reached a composite mean Intersection-over-Union (mIoU) of 69.73% on the official 1,815-image test set. Our model adapts the image encoder of a recent visual foundation model, Segment Anything Model 3 (SAM3), with a lightweight decoder. Beyond this, we contribute two techniques and one empirical finding: (i) a self-distillation scheme that re-uses SAM3 itself, prompted with ground-truth boxes, as a teacher on the classes where it outperforms our own model; (ii) an image-level multi-scale test-time augmentation scheme that restores multi-scale inference for a fixed-input-size model by rescaling the image rather than the model input; and (iii) the finding that an aggressive photometric distortion from a winning 2025 GOOSE 2D entry, transplanted onto our pipeline, is its single largest source of improvement.

cs.CV↗

A Resilience Evaluation Framework for Electric Distribution Systems: Historical Weather Conditioning, Sensitivity Analysis, and a Flooding-Aware Extension

Evaluating resilience in electric distribution systems under severe weather requires models that can connect network topology, hazard simulation, fragility modeling, restoration assumptions, repair strategy, and downstream consequences. This paper extends our prior graph-based resilience evaluation framework for power distribution systems in three ways: it adds analysis conditioned on historical events with real outage and weather data, introduces sensitivity studies for key modeling assumptions, and includes a coupled power-flooding extension for sewage-backup assessment. Historical wind events drive Monte Carlo simulations conditioned on real weather, and the observed outage trajectories are treated as realized historical samples for comparison. Wind-event resilience metrics stabilize at approximately 256 episodes, and outage peak, duration, and outage intensity change systematically with fragility parameters, network topology, restoration assumptions, and repair strategies. In a separate 1000-episode joint power-flooding simulation, episodes with at least one flooded customer occur in 1.9% of episodes overall, and both flood occurrence and flood intensity increase with outage intensity, showing a selective power-to-flood consequence pathway. Overall, the framework provides a practical basis for resilience assessment, comparative scenario analysis, and coupled power-flooding studies in a limited public-data setting, while also suggesting that more detailed utility data could further improve simulation realism.

eess.SY↗

Revisiting the Independence Assumption in LEO Satellite-to-Ground Optical Links: A State-Coupled Joint Fading Model

Performance analysis of low Earth orbit (LEO) satellite-to-ground optical links relies on composite fading models that typically evaluate scintillation and angular loss under the assumption of statistical independence. While ensuring analytical tractability, this assumption decouples fading mechanisms driven by the same atmospheric turbulence and fails to capture the distinct effects of free atmosphere (FA) and boundary layer (BL) perturbations. To model this coupling while preserving tractability, this paper develops a state-coupled joint fading model. In the proposed framework, aperture-averaged scintillation and effective angular loss are jointly characterized by a discrete slow atmospheric state, parameterized by separate FA and BL scaling factors. By replacing unconditional independence with state-conditioned independence, the model enables a closed-form derivation of the outage probability, preserving the computational simplicity of the independent baseline. Numerical results show that the independent baseline can misestimate outage under non-nominal layered turbulence states. This outage prediction bias varies with elevation because the relative roles of scintillation and angular loss change with the link geometry, resulting in different residual angular correction requirements for a given outage target.

eess.SP↗

Tube-Structured Incremental Semantic HARQ for Generative Video Receivers

Generative semantic communication uses receiver-side generative priors to reconstruct visual content from compact semantics, making it attractive for bandwidth-limited multimedia delivery. For video, reliable recovery remains difficult because errors accumulate over time, useful evidence is temporally correlated, and the receiver must make decisions under limited interaction, retransmission, and reconstruction budgets. Existing generative semantic communication studies mainly emphasize representation, compression, or generative reconstruction, while recent error-resilient and semantic-HARQ methods still largely operate on encoder-defined or frame-block retransmission units. This paper studies receiver-driven semantic HARQ for generative video reconstruction under a budget-constrained AoIS-AUC objective and argues that the retransmission primitive is itself an important system design variable. We propose tube-structured package-native requests, in which temporally local packages are the channel-visible HARQ objects and are transmitted, dropped, received, and committed at package granularity. Under a controlled comparison protocol with matched backbone, budgets, and channel model, this primitive yields lower time-weighted recovery cost than competitive block-based baselines in practically relevant moderate-to-harsh regimes, while the gap naturally shrinks in near-clean channels. The gain mainly appears as earlier stabilization of the recovery trajectory, while final-quality endpoints remain broadly comparable, and it persists even against a tube-aware block-ranking baseline.

eess.IV↗

Multi-Scale Wavelet Transformers for Operator Learning of Dynamical Systems

Recent years have seen a surge in data-driven surrogates for dynamical systems that can be orders of magnitude faster than numerical solvers. However, many machine learning-based models such as neural operators exhibit spectral bias, attenuating high-frequency components that often encode small-scale structure. This limitation is particularly damaging in applications such as weather forecasting, where misrepresented high frequencies can induce long-horizon instability. To address this issue, we propose multi-scale wavelet transformers (MSWTs), which learn system dynamics in a tokenized wavelet domain. The wavelet transform explicitly separates low- and high-frequency content across scales. MSWTs leverage a wavelet-preserving downsampling scheme that retains high-frequency features and employ wavelet-based attention to capture dependencies across scales and frequency bands. Experiments on chaotic dynamical systems show substantial error reductions and improved long horizon spectral fidelity. On the ERA5 climate reanalysis, MSWTs further reduce climatological bias, demonstrating their effectiveness in a real-world forecasting setting.

cs.LG↗

Seeing the Evidence, Missing the Answer: Tool-Guided Vision-Language Models on Visual Illusions

Vision-language models (VLMs) exhibit a systematic bias when confronted with classic optical illusions: they overwhelmingly predict the illusion as "real" regardless of whether the image has been counterfactually modified. We present a tool-guided inference framework for the DataCV 2026 Challenge (Tasks I and II) that addresses this failure mode without any model training. An off-the-shelf vision-language model is given access to a small set of generic image manipulation tools: line drawing, region cropping, side-by-side comparison, and channel isolation, together with an illusion-type-routing system prompt that prescribes which tools to invoke for each perceptual question category. Critically, every tool call produces a new, immutable image resource appended to a persistent registry, so the model can reference and compose any prior annotated view throughout its reasoning chain. Rather than hard-coding illusion-specific modules, this generic-tool-plus-routing design yields strong cross-structural generalization: performance remained consistent from the validation set to a test set containing structurally unfamiliar illusion variants (e.g., Mach Bands rotated from vertical to horizontal stacking). We further report three empirical observations that we believe warrant additional investigation: (i) a strong positive-detection bias likely rooted in imbalanced illusion training data, (ii) a striking dissociation between pixel-accurate spatial reasoning and logical inference over self-generated annotations, and (iii) pronounced sensitivity to image compression artifacts that compounds false positives.

cs.CV↗

A Deep Learning-Based Method for Power System Resilience Evaluation

Power system resilience is vital to modern society, as outages caused by extreme weather can severely disrupt communities. Existing statistical and simulation-based methods for resilience quantification are either retrospective or rely on simplified physical models, limiting their applicability. This paper proposes a deep learning-based framework that integrates historical outage and weather data to predict event-level resilience, measured using the resilience trapezoid method. The trained model is then applied to a benchmark weather dataset to estimate regional resilience, with optional socioeconomic and demographic factors incorporated as weighting terms when policymakers wish to emphasize the needs of specific population groups. The effectiveness of the framework is first validated on simulated outage records, showing strong agreement between predicted and simulated resilience values. It is then applied to real historical outage data to assess the resilience of actual power systems. Beyond evaluation, the results can guide targeted investments in distributed energy resources to improve resilience in vulnerable regions.

eess.SY↗

Synthetic Defect Image Generation for Power Line Insulator Inspection Using Multimodal Large Language Models

Utility companies increasingly rely on drone imagery for post-event and routine inspection, but training accurate defect-type classifiers remains difficult because defect examples are rare and inspection datasets are often limited or proprietary. We address this data-scarcity setting by using an off-the-shelf multimodal large language model (MLLM) as a training-free image generator to synthesize defect images from visual references and text prompts. Our pipeline increases diversity via dual-reference conditioning, improves label fidelity with lightweight human verification and prompt refinement, and filters the resulting synthetic pool using an embedding-based selection rule based on distances to class centroids computed from the real training split. We evaluate on ceramic insulator defect-type classification (shell vs. glaze) using a public dataset with a realistic low training-data regime (104 real training images; 152 validation; 308 test). Augmenting the 10% real training set with embedding-selected synthetic images improves test F1 score (harmonic mean of precision and recall) from 0.615 to 0.739 (20% relative), corresponding to an estimated 4--5x data-efficiency gain, and the gains persist with stronger backbone models and frozen-feature linear-probe baselines. These results suggest a practical, low-barrier path for improving defect recognition when collecting additional real defects is slow or infeasible.

cs.CV↗

Diffusion-Aided Bandwidth-Efficient Semantic Communication with Adaptive Requests

Semantic communication focuses on conveying the task-relevant meaning rather than exact bitwise recovery. For image transmission with a generative receiver, relying only on text descriptions can be insufficient to preserve instance-specific visual evidence, whereas sending dense latent representations can incur substantial overhead. This paper presents a receiver-driven closed-loop scheme that transmits a short caption together with an initial sparse subset of latent blocks, and then uses feedback to request additional blocks only when needed. At each round, the receiver reconstructs the image via latent diffusion inpainting and applies a semantic consistency check between a caption generated from the reconstruction and the received caption, using a lightweight language similarity score such as ROUGE-L. The receiver stops early once a target consistency level is met, and otherwise requests a small number of additional latent blocks to refine the reconstruction. Experiments on Flickr30k over AWGN channels demonstrate a controllable rate-quality tradeoff. Adaptive feedback achieves the strongest semantic alignment and the lowest failure rate, outperforming budget-matched one-shot transmission while typically using fewer latent blocks than always-on retransmission.

cs.IT↗

Thompson Sampling in Function Spaces via Neural Operators

We propose an extension of Thompson sampling to optimization problems over function spaces where the objective is a known functional of an unknown operator's output. We assume that queries to the operator (such as running a high-fidelity simulator or physical experiment) are costly, while functional evaluations on the operator's output are inexpensive. Our algorithm employs a sample-then-optimize approach using neural operator surrogates. This strategy avoids explicit uncertainty quantification by treating trained neural operators as approximate samples from a Gaussian process (GP) posterior. We derive regret bounds and theoretical results connecting neural operators with GPs in infinite-dimensional settings. Experiments benchmark our method against other Bayesian optimization baselines on functional optimization tasks involving partial differential equations of physical systems, demonstrating better sample efficiency and significant performance gains.

stat.ML↗

Assessing the Performance and Impact of PV Technologies on Storage in Hybrid Renewable Systems

Traditional monofacial photovoltaic (mPV) systems are commonly adopted and well-documented because of their lower upfront costs in comparison to bifacial photovoltaic (bPV) systems. This study investigates how PV technologies impact energy storage in grid-scale hybrid renewable systems, focusing on optimizing and assessing the performance of mPV and bPV technologies integrated with pumped storage hydropower. Using Ludington City, Michigan as a case study and analyzing realworld data such as solar irradiance, ambient temperature, and utility-scale load profiles, the research highlights the operational and economic benefits of bPV systems. The results reveal that bPV systems can pump approximately 10.38% more water annually to the upper reservoir while achieving a lower levelized cost of energy ($0.0578/kWh for bPV vs. $0.0672/kWh for mPV). This study underscores the outstanding potential of bPV systems in enhancing energy storage and management strategies, contributing to a more sustainable and resilient renewable energy future.

eess.SY↗

Adaptive Lighting Control in Visible Light Systems: An Integrated Sensing, Communication, and Illumination Framework

Indoor visible light communication (VLC) is a promising sixth-generation (6G) technology, as its directional and sensitive optical signals are naturally suited for integrated sensing and communication (ISAC). However, current research mainly focuses on maximizing data rates and sensing accuracy, creating a conflict between high performance, high energy consumption, and user visual comfort. This paper proposes an adaptive integrated sensing, communication, and illumination (ISCI) framework that resolves this conflict by treating energy savings as a primary objective. The framework's mechanism first partitions the receiving plane using a geometric methodology, defining an activity area and a surrounding non-activity area to match distinct user requirements. User location, determined using non-line-of-sight (NLOS) sensing, then acts as a dynamic switch for the system's optimization objective. The system adaptively shifts between minimizing total transmit power while guaranteeing communication and illumination performance in the activity area and maximizing signal-to-noise ratio (SNR) uniformity in the non-activity area. Numerical results confirm that this adaptive ISCI approach achieves 53.59% energy savings over a non-adaptive system and improves SNR uniformity by 57.79%, while satisfying all illumination constraints and maintaining a mean localization error of 0.071 m.

eess.SY↗

NOMA for Visible Light Communications: Recent Advances and Future Directions

Rapidly increasing demand for high speed data is pushing 6G wireless networks to support larger link scales, lower latency, and higher spectral efficiency. Visible light communications (VLC) is a strong complement to radio frequency (RF) systems within 6G. The latest ITU G.9991 and IEEE 802.11bb standards are adapted from cable and RF wireless technologies for use in VLC, so they do not fully exploit the optical nature of light links. VLC links are often asymmetric between uplink and downlink, which makes TDMA style protocols inefficient when many users generate bursty and asymmetric traffic. Compared with RF, the strong directionality and frequent line of sight in VLC can mitigate hidden and exposed terminals, yet these effects can still appear under limited field of view, blockage, or reflections. CSMA/CA and related methods remain usable in VLC and in RF plus VLC networks, but they usually need design tweaks such as RTS/CTS or directional sensing to perform well. Although the optical spectrum is vast, the bandwidth of practical LEDs and of common PIN or APD receivers is limited, so efficient multiple access can yield large gains. This motivates a clean slate design for VLC, especially at the MAC layer. NOMA, first explored in 5G RF systems, is also promising for 6G VLC. It lets multiple users share the same time and frequency resources while tolerating controlled interference. This paper reviews progress in VLC and in NOMA based VLC, outlines key optimization constraints and objectives, surveys scenarios that fit NOMA in VLC, and points to several directions for future work.

eess.SP↗