Semantic communication, augmented by knowledge bases (KBs), offers substantial reductions in transmission overhead and resilience to errors. However, existing methods predominantly rely on end-to-end training to construct KBs, often failing to fully capitalize on the rich information available at communication devices. Motivated by the growing convergence of sensing and communication, we introduce a novel Position-Aided Semantic Communication (PASC) framework, which integrates localization into semantic transmission. This framework is particularly designed for position-based image communication, such as real-time uploading of outdoor camera-view images. By utilizing the position, the framework retrieves corresponding maps, and then an advanced foundation model (FM)-driven view generator is employed to synthesize images closely resembling the target images. The PASC framework further leverages the FM to fuse the synthesized image with deviations from the real one, enhancing semantic reconstruction. Notably, the framework is highly flexible, capable of adapting to dynamic content and fluctuating channel conditions through a novel FM-based parameter optimization strategy. Additionally, the challenges of real-time deployment are addressed, with the development of a hardware testbed to validate the framework. Simulations and real-world tests demonstrate that the proposed PASC approach not only significantly boosts transmission efficiency, but also remains robust in diverse and evolving transmission scenarios.

Foundation model-based semantic transmission has recently shown great potential in wireless image communication. However, existing methods exhibit two major limitations: (i) they overlook the varying importance of semantic components for specific downstream tasks, and (ii) they insufficiently exploit wireless domain knowledge, resulting in limited robustness under dynamic channel conditions. To overcome these challenges, this paper proposes a foundation model-based adaptive semantic image transmission system for dynamic wireless environments, such as autonomous driving. The proposed system decomposes each image into a semantic segmentation map and a compressed representation, enabling task-aware prioritization of critical objects and fine-grained textures. A task-adaptive precoding mechanism then allocates radio resources according to the semantic importance of extracted features. To ensure accurate channel information for precoding, a channel estimation knowledge map (CEKM) is constructed using a conditional diffusion model that integrates user position, velocity, and sparse channel samples to train scenario-specific lightweight estimators. At the receiver, a conditional diffusion model reconstructs high-quality images from the received semantic features, ensuring robustness against channel impairments and partial data loss. Simulation results on the BDD100K dataset with multi-scenario channels generated by QuaDRiGa demonstrate that the proposed method outperforms existing approaches in terms of perceptual quality (SSIM, LPIPS, FID), task-specific accuracy (IoU), and transmission efficiency. These results highlight the effectiveness of integrating task-aware semantic decomposition, scenario-adaptive channel estimation, and diffusion-based reconstruction for robust semantic transmission in dynamic wireless environments.

Foundation models (FMs) unlock unprecedented multimodal and multitask intelligence, yet their cloud-centric deployment precludes real-time responsiveness and compromises user privacy. Meanwhile, monolithic execution at the edge remains infeasible under stringent resource limits and uncertain network dynamics. To bridge this gap, we propose a microservice-based FM inference framework that exploits the intrinsic functional asymmetry between heavyweight core services and agile light services. Our two-tier deployment strategy ensures robust Quality of Service (QoS) under resource contention. Specifically, core services are placed statically via a long-term network-aware integer program with sparsity constraints to form a fault-tolerant backbone. On the other hand, light services are orchestrated dynamically by a low-complexity online controller that integrates effective capacity theory with Lyapunov optimization, providing probabilistic latency guarantees under real-time workload fluctuations. Simulations demonstrate that our framework achieves over 84

We propose TC-AE, a ViT-based architecture for deep compression autoencoders. Existing methods commonly increase the channel number of latent representations to maintain reconstruction quality under high compression ratios. However, this strategy often leads to latent representation collapse, which degrades generative performance. Instead of relying on increasingly complex architectures or multi-stage training schemes, TC-AE addresses this challenge from the perspective of the token space, the key bridge between pixels and image latents, through two complementary innovations: Firstly, we study token number scaling by adjusting the patch size in ViT under a fixed latent budget, and identify aggressive token-to-latent compression as the key factor that limits effective scaling. To address this issue, we decompose token-to-latent compression into two stages, reducing structural information loss and enabling effective token number scaling for generation. Secondly, to further mitigate latent representation collapse, we enhance the semantic structure of image tokens via joint self-supervised training, leading to more generative-friendly latents. With these designs, TC-AE achieves substantially improved reconstruction and generative performance under deep compression. We hope our research will advance ViT-based tokenizer for visual generation.

Mutual information (MI)-based guidelines have recently proven to be effective for designing task-oriented communication systems, where the ultimate goal is to extract and transmit task-relevant information for downstream tasks. This paper provides a comprehensive overview of MI-empowered task-oriented communication, highlighting how MI-based methods can serve as a unifying design framework in various task-oriented communication scenarios. We begin with the roadmap of MI for designing task-oriented communication systems, and then introduce the roles and applications of MI to guide feature encoding, transmission optimization, and efficient training with two case studies. We further provide a deeper examination of the limitations and challenges associated with MI-based methods. Lastly, we highlight several open issues in MI-based task-oriented communication to inspire future research.

Inter-agent communication serves as an effective mechanism for enhancing performance in collaborative multi-agent reinforcement learning (MARL) systems. However, the inherent communication latency in practical systems induces both action decision delays and outdated information sharing, impeding MARL performance gains, particularly in time-critical applications like autonomous driving. In this work, we propose a Value-of-Information aware Low-latency Communication (VIL2C) scheme that proactively adjusts the latency distribution to mitigate its effects in MARL systems. Specifically, we define a Value of Information (VoI) metric to quantify the importance of delayed messages on the recipient agent's decision. We then design a VoI aware resource allocation method that dynamically prioritizes message transmission based on each delayed message's importance. Moreover, we propose a progressive message reception mechanism to adaptively adjust the reception duration based on received messages. We derive the optimized VoI aware resource allocation and theoretically prove the performance advantage of the proposed VIL2C scheme. Extensive experiments demonstrate that VIL2C outperforms existing approaches under various communication conditions. These gains are attributed to the low-latency transmission of high-VoI messages via resource allocation and the elimination of unnecessary waiting periods via adaptive reception duration.

Recent advancements in foundation models (FMs) have attracted increasing attention in the wireless communication domain. Leveraging the powerful multi-task learning capability, FMs hold the promise of unifying multiple tasks of wireless communication with a single framework. with a single framework. Nevertheless, existing wireless FMs face limitations in the uniformity to address multiple tasks with diverse inputs/outputs across different communication scenarios.In this paper, we propose a MUlti-taSk Environment-aware FM (MUSE-FM) with a unified architecture to handle multiple tasks in wireless communications, while effectively incorporating scenario information.Specifically, to achieve task uniformity, we propose a unified prompt-guided data encoder-decoder pair to handle data with heterogeneous formats and distributions across different tasks. Besides, we integrate the environmental context as a multi-modal input, which serves as prior knowledge of environment and channel distributions and facilitates cross-scenario feature extraction. Simulation results illustrate that the proposed MUSE-FM outperforms existing methods for various tasks, and its prompt-guided encoder-decoder pair improves the scalability for new task configurations. Moreover, the incorporation of environment information improves the ability to adapt to different scenarios.

Fine-tuning large language models (LLMs) on custom datasets has become a standard approach for adapting these models to specific domains and applications. However, recent studies have shown that such fine-tuning can lead to significant degradation in the model's safety. Existing defense methods operate at the sample level and often suffer from an unsatisfactory trade-off between safety and utility. To address this limitation, we perform a systematic token-level diagnosis of safety degradation during fine-tuning. Based on this, we propose token-level data selection for safe LLM fine-tuning (TOSS), a novel framework that quantifies the safety risk of each token by measuring the loss difference between a safety-degraded model and a utility-oriented model. This token-level granularity enables accurate identification and removal of unsafe tokens, thereby preserving valuable task-specific information. In addition, we introduce a progressive refinement strategy, TOSS-Pro, which iteratively enhances the safety-degraded model's ability to identify unsafe tokens. Extensive experiments demonstrate that our approach robustly safeguards LLMs during fine-tuning while achieving superior downstream task performance, significantly outperforming existing sample-level defense methods.

MeanFlow generators achieve fast few-step sampling by predicting average velocities over time intervals, making them attractive for efficient generation. Reinforcement learning (RL) has become a powerful way to align diffusion and flow models with human preferences and task-specific objectives. In particular, DiffusionNFT offers an efficient forward-process RL framework that does not require reverse-process trajectories or likelihood estimation. However, applying such RL methods to MeanFlow remains underexplored. DiffusionNFT optimizes instantaneous velocities, whereas MeanFlow samples with average velocities. To bridge this gap, we introduce MeanFlowNFT. Inspired by the MeanFlow identity, which bridges average and instantaneous velocities, we construct an induced instantaneous-velocity predictor. We apply the DiffusionNFT objective to this predictor, making reward optimization well-defined for MeanFlow. Sampling remains based on the average velocity, preserving MeanFlow's fast few-step generation. We further prove that MeanFlowNFT inherits DiffusionNFT's strict policy-improvement guarantee. Experiments on image and video generation show that MeanFlowNFT consistently improves baselines. Moreover, it outperforms prior state-of-the-art RL-tuned few-step generators on most metrics (6 of 8 on SD3.5-M), and can even surpass multi-step RL-tuned diffusion while using only a few sampling steps. For instance, on Wan 2.1, 4-step MeanFlowNFT reaches a VBench score of 84.33, surpassing 50-step LongCat-Video RL (82.57).

Recent advancements in large language models (LLMs) have been largely driven by vast amounts of training data. As public data are exhausted, fine-tuning LLMs with private data has become increasingly important. However, due to privacy concerns associated with sensitive data, traditional centralized training approaches are not feasible. Federated learning (FL) offers a promising solution, as it enables collaborative model training across multiple mobile devices without data exchange. Nevertheless, it is challenging to fine-tune LLMs with FL on resource-constrained and heterogeneous clients. To reduce the cost, we employ Low-Rank Adaptation (LoRA) for efficient fine-tuning. Existing federated LoRA fine-tuning methods suffer from inaccurate aggregation, and the resulting noise may lead to performance loss that may be more severe for LLMs. In this paper, we propose a novel method, FedOTAB, to mitigate this issue. FedOTAB first correctly aggregates all local LoRA modules and then fixes the matrix $\bf {A}$ or $\bf {B}$ to optimize the corresponding best approximate matrix $\bf {B}^*$ or $\bf {A}^*$ for each local LoRA module by minimizing the induced noise. In addition, we design a mechanism to freeze the matrices $\bf {A}$ and $\bf {B}$ alternately to stabilize optimization. By transmitting only the updated part, that is, half of the LoRA module, FedOTAB enjoys favorable communication efficiency. By explicitly splitting LoRA into local and global parts, FedOTAB effectively integrates both local and global knowledge. Extensive experiments demonstrate that FedOTAB excels in both homogeneous and heterogeneous rank scenarios, consistently outperforms state-of-the-art methods with remarkably low communication and computational overhead. We hope that this work provides valuable insights for federated LoRA fine-tuning, offering a practical and scalable solution to fine-tuning large models in mobile environments.

As intents unfold and environments change, multi-turn agents face continuously shifting decision contexts. Although reusing past experience is intuitively appealing, existing approaches remain limited: full trajectories are often too context-specific to transfer, while tool-level reuse ignores the context and environment. In this paper, we introduce a hybrid episodic–procedural memory strategy (H-EPM) that enables experience-induced self-evolution of multi-turn tool-use policies, by adaptively reusing partially overlapping successful experiences in both inference and training. Inspired by human episodic–procedural integration, we build a tool graph from accumulated trajectories, where recurring tool-to-tool dependencies capture procedural routines and each edge is augmented with a compact episodic summaries of relevant context. At inference, the agent dynamically balances episodic recall for contextual reasoning and procedural execution for routine steps. Beyond inference, H-EPM introduces a memory-guided reinforcement learning paradigm that directly addresses a core challenge in multi-turn agent RL: ineffective exploration over long trajectories. By biasing exploration toward historically successful tool transitions, H-EPM learns a stronger policy that generalizes during inference without relying on domain-specific experience collection. Experiments show that H-EPM consistently delivers substantial inference-time gains over strong baselines across multi-turn tool-use benchmarks, reaching up to 50\%+. It also boosts RL policy performance, achieving up to 40\%+ improvement on out-of-distribution tasks.

As a mainstream technique for 3D reconstruction, 3D Gaussian splatting (3DGS) has been applied in a wide range of applications and services. Recent studies have revealed critical vulnerabilities in this pipeline and introduced computation cost attacks that lead to malicious resource occupancies and even denial-of-service (DoS) conditions, thereby hindering the reliable deployment of 3DGS. In this paper, we propose the first effective and comprehensive black-box defense framework, named RemedyGS, against such computation cost attacks, safeguarding 3DGS reconstruction systems and services. Our pipeline comprises two key components: a detector to identify the attacked input images with poisoned textures and a purifier to recover the benign images from their attacked counterparts, mitigating the adverse effects of these attacks. Moreover, we incorporate adversarial training into the purifier to enforce distributional alignment between the recovered and original natural images, thereby enhancing the defense efficacy. Experimental results demonstrate that our framework effectively defends against white-box, black-box, and adaptive attacks in 3DGS systems, achieving state-of-the-art performance in both safety and utility.

Wireless networks are evolving from connectivity-oriented infrastructures into intelligent and personalized service platforms. Existing wireless intelligence remains centered on network-side optimization, improving objectives such as throughput, latency, and coverage. Nevertheless, besides network performance, wireless intelligence also depends on user-perceived experience via application context, mobility routine, service cost, privacy preference, and long-term usage behavior. This article proposes WISPA, a Wireless Intelligent Self-evolving Personal Agent framework for automated terminal-side resource management based on large language model (LLM)-based agent. To overcome the resource constraints on terminals, WISPA decouples the latency-sensitive online resource execution from offline LLM agent reflection. In this way, a lightweight online executor makes deterministic resource decisions using interpretable preference parameters; While an offline LLM agent analyzes terminal-side traces, refines user profiles, and updates online preference parameters for subsequent decisions. At last, we demonstrate the practical applicability and benefits of WISPA for terminal-side resource allocations on a campus commute route. Numerical results show that WISPA learns user-specific connection styles and adapts access decisions as preferences change.

In image editing, it is essential to incorporate a context image to convey the user's precise requirements, such as subject appearance or image style. Existing training-based visual context-aware editing methods incur data collection effort and training cost. On the other hand, the training-free alternatives are typically established on diffusion inversion, which struggles with consistency and flexibility. In this work, we propose VicoEdit, a training-free and inversion-free method to inject the visual context into the pretrained text-prompted editing model. More specifically, VicoEdit directly transforms the source image into the target one based on the visual context, thereby eliminating the need for inversion that can lead to deviated trajectories. Moreover, we design a posterior sampling approach guided by concept alignment to enhance the editing consistency. Empirical results demonstrate that our training-free method achieves even better editing performance than the state-of-the-art training-based models.

Mixture-of-Experts (MoE) Multimodal large language models (MLLMs) excel at vision-language tasks, but they suffer from high computational inefficiency. To reduce inference overhead, expert skipping methods have been proposed to deactivate redundant experts based on the current input tokens. However, we find that applying these methods-originally designed for unimodal large language models (LLMs)-to MLLMs results in considerable performance degradation. This is primarily because such methods fail to account for the heterogeneous contributions of experts across MoE layers and modality-specific behaviors of tokens within these layers. Motivated by these findings, we propose MoDES, the first training-free framework that adaptively skips experts to enable efficient and accurate MoE MLLM inference. It incorporates a globally-modulated local gating (GMLG) mechanism that integrates global layer-wise importance into local routing probabilities to accurately estimate per-token expert importance. A dual-modality thresholding (DMT) method is then applied, which processes tokens from each modality separately, to derive the skipping schedule. To set the optimal thresholds, we introduce a frontier search algorithm that exploits monotonicity properties, cutting convergence time from several days to a few hours. Extensive experiments for 3 model series across 13 benchmarks demonstrate that MoDES far outperforms previous approaches. For instance, when skipping 88

This paper presents WorldPlay, a streaming video diffusion model that enables real-time, interactive world modeling with long-term geometric consistency, resolving the trade-off between speed and memory that limits current methods. WorldPlay draws power from three key innovations. 1) We use a Dual Action Representation to enable robust action control in response to the user's keyboard and mouse inputs. 2) To enforce long-term consistency, our Reconstituted Context Memory dynamically rebuilds context from past frames and uses temporal reframing to keep geometrically important but long-past frames accessible, effectively alleviating memory attenuation. 3) We also propose Context Forcing, a novel distillation method designed for memory-aware model. Aligning memory context between the teacher and student preserves the student's capacity to use long-range information, enabling real-time speeds while preventing error drift. Taken together, WorldPlay generates long-horizon streaming 720p video at 24 FPS with superior consistency, comparing favorably with existing techniques and showing strong generalization across diverse scenes.

Beamforming plays a key role in multiple-input-multiple-output (MIMO) communication systems. However, conventional beamforming design normally requires accurate instantaneous channel state information (CSI) and iterative optimization, which incur substantial pilot overhead and computational complexity. Recognizing that radio propagation is intrinsically governed by the physical geometry, we develop a 3D Gaussian splatting for environment-aware beamforming (GSBF) pipeline based on multi-modal data, which characterizes the environment through a persistent 3D Gaussian representation. Specifically, GSBF models the environmental scattering response with reciprocity-preserving bidirectional spherical Gaussian (Bi-SG) kernels and performs two-sided electromagnetic rasterization to render an angular propagator map. The rendered map is then aggregated through an over-complete array-manifold dictionary and projected to the constant-modulus beamformers, thereby synthesizing beams directly from the access point (AP) pose and user position without online instantaneous CSI. Simulations demonstrate that GSBF consistently outperforms baselines such as exhaustive beam alignment (EBA) with lower latency.

Offline Safe Reinforcement Learning (OSRL) aims to learn a policy that achieves high performance in sequential decision-making while satisfying safety constraints, using only pre-collected datasets. Recent works, inspired by the strong capabilities of Generative Models (GMs), reformulate decision-making in OSRL as a conditional generative process, where GMs generate desirable actions conditioned on predefined reward and cost return-to-go values. However, GM-assisted methods face two major challenges in constrained settings: (1) they lack the ability to ``stitch'' optimal transitions from suboptimal trajectories within the dataset, and (2) they struggle to balance reward maximization with constraint satisfaction, particularly when tested with imbalanced human-specified reward-cost conditions. To address these issues, we propose Goal-Assisted Stitching (GAS), a novel algorithm designed to enhance stitching capabilities while effectively balancing reward maximization and constraint satisfaction. To enhance the stitching ability, GAS first augments and relabels the dataset at the transition level, enabling the construction of high-quality trajectories from suboptimal ones. GAS also introduces novel goal functions, which estimate the optimal achievable reward and cost goals from the dataset. These goal functions, trained using expectile regression on the relabeled and augmented dataset, allow GAS to accommodate a broader range of reward-cost return pairs and achieve a better tradeoff between reward maximization and constraint satisfaction compared to human-specified values. The estimated goals then guide policy training, ensuring robust performance under constrained settings. Furthermore, to improve training stability and efficiency, we reshape the dataset to achieve a more uniform reward-cost return distribution. Empirical results validate the effectiveness of GAS, demonstrating superior performance in balancing reward maximization and constraint satisfaction compared to existing methods.

The expanding application of smart sensing has created a growing demand for the accurate understanding of human action at the network edge. Traditional approaches require massive video data to be transmitted from resource-constrained edge devices to powerful cloud servers, incurring prohibitive uplink bandwidth consumption and unacceptable latency while raising privacy concerns. To overcome these bottlenecks, we propose a task-oriented communication framework for human action understanding (TOAU) through edge-cloud collaboration. Our framework utilizes a monocular pose estimator to extract continuous joint coordinates from raw videos, followed by a vector quantized variational autoencoder (VQ-VAE) to convert these coordinates into discrete motion tokens. Consequently, only a compact sequence of codebook indices is transmitted over the network, consuming as few as 9 bits per frame and avoiding privacy leakages. At the cloud server, a lightweight projector aligns these motion tokens with the embedding space of a large vision-language model (VLM) to facilitate complex action understanding, which is trained with an efficient instruction tuning paradigm. Comprehensive evaluations on three benchmarks demonstrate that our TOAU system reduces the transmission payload to approximately 1% and the system latency to around 20% compared to video codec-based solutions, while delivering comparable action understanding accuracy.

LLM agents are emerging as a key enabler for autonomous wireless network management. Reliably deploying them, however, demands benchmarks that reflect real engineering risk. Existing wireless benchmarks evaluate single isolated capabilities and treat all errors uniformly, missing both cascaded-chain failures and catastrophic unit confusions (e.g., dB vs. dBm). We present , the first tolerance-aware, tool-integrated benchmark for LLM-based wireless agents. is organized as a three-tier cognitive hierarchy: domain knowledge reasoning (WCHW, 1,392 items), intent-driven resource allocation (WCNS, 1,000 items), and proactive multi-step decisions under mobility (WCMSA, 1,000 items). Moreover, is established on three design principles: (i) tolerance-aware scoring with catastrophic-error detection; (ii) tool-necessary tasks requiring a 3GPP-compliant ray-tracing query for channel quality; and (iii) Chain-of-Thought (CoT)-traceable items, where every benchmark item ships with a complete CoT trajectory enabling fine-grained diagnosis of where in the reasoning chain an agent fails. Our numerical results show that the direct-prompting model (GPT-4o) scores 68%, trailing a tool-integrated agent (84.64%) by 16.64 pp; 23% of errors are catastrophic failures invisible to exact-match metrics. More importantly, the hierarchy decomposes errors into four actionable diagnostic categories that flat evaluation cannot reveal. Code and data: https://wirelessbench.github.io/.