Abstract Crowdsensing networks face complex security threats due to their open architecture and distributed collaboration model. However, existing defense methods neglect the heterogeneity of sensing nodes and information asymmetry between attackers and defenders, resulting in traditional static defense strategies that struggle to adapt to dynamic adversarial environments. To address this challenge, this paper proposes a dynamic collaborative defense model driven by incomplete information differential games. Firstly, a dynamic differential equation system with random perturbation factors is established to characterize the security state evolution process of sensing nodes. Subsequently, an incomplete information differential game model is constructed to analyze attack and defense behaviors, with a Bayesian strategy inference method designed to infer potential strategy preferences of attackers through real-time attack-defense interaction results and node states. This approach overcomes the limitations of static belief updates in traditional incomplete information games. Finally, a defense decision-making algorithm is designed based on optimal control theory. Simulation results demonstrate that the proposed model accurately characterizes strategic interactions between attackers and defenders while providing real-time defensive strategies based on the inference method, improving defense success rates by 24.2%. Additionally, the model exhibits strong adaptability in dynamic network environments, enhancing network availability by 22.3%.

World models have demonstrated superiority in autonomous driving, particularly in the generation of multi-view driving videos. However, significant challenges still exist in generating customized driving videos. In this paper, we propose DriveDreamer-2, which incorporates a Large Language Model (LLM) to facilitate the creation of user-defined driving videos. Specifically, a trajectory generation function library is developed to produce trajectories that conform to user descriptions. Subsequently, an HDMap generator is designed to learn the mapping from trajectories to road structures. Ultimately, we propose the Unified Multi-View Model (UniMVM) to enhance temporal and spatial coherence in the generated multi-view driving videos. To the best of our knowledge, DriveDreamer-2 is the first world model to generate customized driving videos, and it can generate uncommon driving videos (e.g., vehicles abruptly cut in) in a user-friendly manner. Besides, experimental results demonstrate that the generated videos enhance the training of driving perception methods (e.g., 3D detection and tracking). Furthermore, video generation quality of DriveDreamer-2 surpasses other state-of-the-art methods, showcasing FID and FVD scores of 11.2 and 55.7, representing relative improvements of ~30% and ~50%.

Closed-loop simulation is crucial for end-to-end autonomous driving. Existing sensor simulation methods (e.g., NeRF and 3DGS) reconstruct driving scenes based on conditions that closely mirror training data distributions. However, these methods struggle with rendering novel trajectory, such as lane changes. Recent works have demonstrated that integrating world model knowledge alleviates these issues. Despite their efficiency, these approaches still encounter difficulties in the accurate representation of more complex maneuvers, with multi-lane shifts being a notable example. Therefore, we introduce ReconDreamer, which enhances driving scene reconstruction through incremental integration of world model knowledge. Specifically, DriveRestorer is proposed to mitigate artifacts via online restoration. This is complemented by a progressive data update strategy designed to ensure high-quality rendering for more complex maneuvers. To the best of our knowledge, ReconDreamer is the first method to effectively render in large maneuvers. Experimental results demonstrate that ReconDreamer outperforms Street Gaussians in the NTA-IoU, NTL-IoU, and FID, with relative improvements by 24.87%, 6.72%, and 29.97%. Furthermore, ReconDreamer surpasses DriveDreamer4D with PVG during large maneuver rendering, as verified by a relative improvement of 195.87% in the NTA-IoU metric and a user study.

Human-motion video generation has been a challenging task, primarily due to the difficulty inherent in learning human body movements. While some approaches have attempted to drive human-centric video generation explicitly through pose control, these methods typically rely on poses derived from existing videos, thereby lacking flexibility. To address this, we propose HumanDreamer, a decoupled human video generation framework that first generates diverse poses from text prompts and then leverages these poses to generate human-motion videos. Specifically, we propose MotionVid, the largest dataset for human-motion pose generation. Based on the dataset, we present MotionDiT, which is trained to generate structured human-motion poses from text prompts. Besides, a novel LAMA loss is introduced, which together contribute to a significant improvement in FID by 62.4%, along with respective enhancements in R-precision for top1, top2, and top3 by 41.8%, 26.3%, and 18.3%, thereby advancing both the Text-to-Pose control accuracy and FID metrics. Our experiments across various Pose-to-Video baselines demonstrate that the poses generated by our method can produce diverse and high-quality human-motion videos. Furthermore, our model can facilitate other downstream tasks, such as pose sequence prediction and 2D-3D motion lifting.

Closed-loop simulation is essential for advancing end-to-end autonomous driving systems. Contemporary sensor simulation methods, such as NeRF and 3DGS, rely predominantly on conditions closely aligned with training data distributions, which are largely confined to forward-driving scenarios. Consequently, these methods face limitations when rendering complex maneuvers (e.g., lane change, acceleration, deceleration). Recent advancements in autonomous-driving world models have demonstrated the potential to generate diverse driving videos. However, these approaches remain constrained to 2D video generation, inherently lacking the spatiotemporal coherence required to capture intricacies of dynamic driving environments. In this paper, we introduce DriveDreamer4D, which enhances 4D driving scene representation leveraging world model priors. Specifically, we utilize the world model as a data machine to synthesize novel trajectory videos, where structured conditions are explicitly leveraged to control the spatial-temporal consistency of traffic elements. Besides, the cousin data training strategy is proposed to facilitate merging real and synthetic data for optimizing 4DGS. To our knowledge, DriveDreamer4D is the first to utilize video generation models for improving 4D reconstruction in driving scenarios. Experimental results reveal that DriveDreamer4D significantly enhances generation quality under novel trajectory views, achieving a relative improvement in FID by 32.1%, 46.4%, and 16.3% compared to PVG, S3Gaussian, and Deformable-GS. Moreover, DriveDreamer4D markedly enhances the spatiotemporal coherence of driving agents, which is verified by a comprehensive user study and the relative increases of 22.6%, 43.5%, and 15.6% in the NTA-IoU metric.

Interactive 3D generation is gaining momentum and capturing extensive attention for its potential to create immersive virtual experiences. However, a critical challenge in current 3D generation technologies lies in achieving real-time interactivity. To address this issue, we introduce WonderTurbo, the first real-time interactive 3D scene generation framework capable of generating novel perspectives of 3D scenes within 0.72 seconds. Specifically, WonderTurbo accelerates both geometric and appearance modeling in 3D scene generation. In terms of geometry, we propose StepSplat, an innovative method that constructs efficient 3D geometric representations through dynamic updates, each taking only 0.26 seconds. Additionally, we design QuickDepth, a lightweight depth completion module that provides consistent depth input for StepSplat, further enhancing geometric accuracy. For appearance modeling, we develop FastPaint, a 2-steps diffusion model tailored for instant inpainting, which focuses on maintaining spatial appearance consistency. Experimental results demonstrate that WonderTurbo achieves a remarkable 15X speedup compared to baseline methods, while preserving excellent spatial consistency and delivering high-quality output.

Video generation has emerged as a promising tool for world simulation, leveraging visual data to replicate real-world environments. Within this context, egocentric video generation, which centers on the human perspective, holds significant potential for enhancing applications in virtual reality, augmented reality, and gaming. However, the generation of egocentric videos presents substantial challenges due to the dynamic nature of first-person viewpoints, the intricate diversity of actions, and the complex variety of scenes encountered. Existing datasets are inadequate for addressing these challenges effectively. To bridge this gap, we present EgoVid-5M, the first high-quality dataset specifically curated for egocentric video generation. EgoVid-5M encompasses over 5 million egocentric video clips and is enriched with detailed action annotations, including fine-grained kinematic control and high-level textual descriptions. To ensure the integrity and usability of the dataset, we implement a sophisticated data cleansing pipeline designed to maintain frame consistency, action coherence, and motion smoothness under egocentric conditions. Furthermore, we introduce EgoDreamer, which is capable of generating egocentric videos driven simultaneously by action descriptions and kinematic control signals. The EgoVid-5M dataset, associated action annotations, and all data cleansing metadata will be released for the advancement of research in egocentric video generation.

Mobile crowdsensing (MCS) is a sensing paradigm that leverages the collaboration of mobile devices to collect and process data, and it has been widely applied in domains such as transportation, environmental monitoring, and public safety. However, its open nature and reliance on cooperative networks render it vulnerable to threats such as malicious nodes, link attacks, and data tampering. Given that MCS systems heavily depend on user participation and a stable network topology, these threats are not only difficult to fully mitigate but may also lead to service interruptions, degradation in data quality, or even system-wide failures—particularly in resource-constrained or unstable network environments. To address these challenges, a method to enhance the survivability of MCS systems is proposed. Firstly, to capture the complex interaction patterns among sensing devices in an MCS system, a neighbor interaction encoding scheme is proposed. This scheme learns the temporal node sensing representation of the dynamic graph by considering not only the information of nodes, edges, and time intervals but also the interactions between neighboring devices. Then, by integrating node-specific attribute information with betweenness centrality, the goal is to predict the importance of nodes in the dynamic graph and classify their roles accordingly. Finally, hierarchical threat response strategies are devised for nodes of varying importance, with adaptive topology reconstruction then performed based on predictions of future links between devices. Simulation experiments were conducted on eight real-world datasets, where the average AP and AUC-ROC under three sampling strategies were 88.84 and 87.86, respectively. These results represent average improvements of 6.24% and 4.35% compared to the baseline methods. Additionally, the effectiveness of the proposed method was validated through a mitigation effect evaluation experiment.

Existing task allocation methods do not consider the uneven distribution of sensing users and ignore the sparse sensing region. This leads to the problem of low sensing coverage of the whole sensing task. In this paper, a method of sparse region prediction and interpretability analysis for mobile crowdsensing is proposed. Firstly, the spatio-temporal graph neural network is used to establish a prediction model for the sensing user’s flow distribution. The model considers the spatial dependency and the influence of sensing user flow on the long-term, medium-term, and short-term prediction results. Next, the prediction results are interpreted in terms of both edges and nodes using GNNExplainer. Then, the sensing range is divided into sparse and non-sparse sensing regions based on the prediction results. Finally, the sensing coverage is maximized with reduced movement. Experimental results under different datasets show that our method effectively reduces movement costs and improves sensing coverage compared to other baseline methods.

Mobile Crowd Sensing is a human-based data collection model, and the approach taken to recommend data collection tasks to users in order to maximize task acceptance rates is an important part of this research. Existing task recommendation methods are based only on intuitive data for unconscious analysis and decision-making, and lack the embodiment of cognitive intelligence. To address the above problem, a conscious task recommendation based on cognitive reasoning computing in Mobile Crowd Sensing has been proposed, using knowledge from cognitive science to simulate the human thinking process in order to achieve warm learning and conscious recommendation of sensing tasks. First, the task attributes are segmented into positive and negative attributes using a Kernel Density Estimation method based on bandwidth self-selection. Then, the user's attribute preferences are diagnosed by the Cognitive Diagnostic Method to obtain the user's preference vector. Finally, get the overall preference trend of users based on the Drift Diffusion Model, and make decisions according to whether the current task drift direction is consistent with the user preference trend. Simulation experiments were conducted using the Taobao dataset, MTurk dataset, and synthetic dataset, it was ultimately proven that conscious task recommendation combined with user cognitive ability effectively reduced RMSE and improved task acceptance rate. RMSE was 10.5%∼70.8% lower than other methods, and the task acceptance rate was basically over 80%, with most of the results being over 90%.

Mobile crowdsensing (MCS) is a combination of crowdsourcing ideas and mobile sensing devices, designed to enable rational allocation of resources at scale. However, the MCS platform is highly vulnerable to injection attacks from malicious participants and fake tasks that interfere with platform service capabilities and sensing activities. To this end, the participant and task submission process is modeled as a multivariate time series, and a detection model for malicious participants and fake tasks (MP-FTD) with a Gaussian prior on the attentional mechanism and a two-stage adversarial training process is proposed. The attention mechanism was corrected using Gaussian bias, and then the corrected attention mechanism was used to obtain the correlation discrepancies between the data. Using the adversarial training method of Generative Adversarial Networks (GAN), the output of the correlation discrepancy reconstruction phase is transformed into a focus score, to amplify the reconstruction error in the output of the focus score reconstruction phase, and to improve the differentiation between the injected data and normal data of malicious attackers. The detection of these malicious attackers will effectively improve the robustness of the sensing platform. Experiments on six real-world datasets showed that the average F1-score reached 93.44%, outperforming the current baseline method, and resulting in an average 12.07% improvement in participant assignment accuracy and an average 12.25% improvement in task assignment accuracy in task assignment experiments.

In the decision-making process of the behavior of mobile crowdsensing, using a single view to learn a user's preference will lead to a mismatch between the user's wishes and the final task recommendation list, resulting in the low efficiency of the model recommendation. Aiming at the lack of perceptual representation and cognitive fusion of multimodal coupled information, a task recommendation method based on heterogeneous multimodal features and decision fusion is proposed. According to the content characteristics of multi-source data in the user's historical task set, several task-task similarity matrices are constructed to align feature dimensions and feature semantics. Using the improved similarity network fusion algorithm, networks composed of multiple content similarity matrices are effectively fused into a similarity network. Considering the influence of the time factor, the tasks that have had interest drift are filtered out from the set of tasks that the user has participated in. Finally, the updated similarity network is clustered to predict the current preference of the user for new tasks. Experimental results based on simulation and real datasets show that the proposed method can effectively improve the accuracy and efficiency of task assignments while improving user satisfaction.

To solve the problem of sensing data redundancy and missing data caused by the uneven distribution of resources in mobile crowd sensing, a two-phased participant selection method based on partial transfer learning is proposed. Firstly, the data is preprocessed. On the one hand, the sensing task features are extracted to analyze the correlation between the source task and the target task feature space. On the other hand, users are divided into active users and passive users according to the historical movement law of sensing users. Secondly, the participant selection in the first stage is carried out. According to the similarity of feature space between the source task and the target task, part of the user resources of the source task are migrated to the target subtask with a similar distribution of its feature space. As a result, the target task can select participants efficiently and accurately. Finally, the participant selection in the second stage is carried out. For the target subtask not covered, the passive users in the subtask area are taken as the assignment object. Simulation results based on real data sets show that this method can effectively improve the task coverage and reduce the perceived excitation cost.

Aiming at the problem that the mobile crowdsensing (MCS) system relies on a specific platform with a large user group presupposed, this paper proposes a sensing user diffusion analysis and incentive method based on knowledge graph reasoning. We consider motivating users to participate under the constraint of limited budget so that the platform and users can get the most benefits. In this paper, we focus on socially aware users represented by self-organizing social networks, combine the knowledge graph to establish a knowledge graph for the crowdsensing system, use rules to derive user influence, and optimize user contributions. With the goal of maximizing social welfare, we propose a social awareness reverse auction (SARA) mechanism, in which the total contribution of users is the key to select winners, and the winners are paid based on critical prices. Through experimental simulations, we verify that SARA is close to the optimal social welfare under budget constraints.

To enhance system survivability when the original mapping scheme fails caused by the change of underlying infrastructures, a dynamic resource virtualisation method based on software definition network (SDN) is proposed. First of all, modelling and formalisation of dynamic resource remapping under SDN environment are constructed, which make the entity resources controlled in the network. Then, resource constraints and evaluation objectives of mapping are considered, and the remapping procedures of nodes and links are described in the case of failure of physical resources. During the process, the detachment factor is introduced as the index to evaluate whether the virtual nodes and links need to be separated, and further be allocated by the greedy algorithm reasonably and dynamically, to realise the virtual network requests. Finally, the simulation experiments are carried out and show that the proposed method is superior to other related methods in virtual request acceptance rate, resource utilisation and average waiting time of service; as a result, improving survivability under the condition that the underlying physical resources change in real time.

Survivability has emerged as a new phase for the development of network security technique, and how to improve the system survivability using effective strategy is an important problem. In this paper, by analyzing the fuzzy matrix game (FMG) theory and network survivability mechanism, a novel strategy selection model for network survivability based on FMG theory and its dynamic analysis method are presented from the macroscopically view. The result of instance analysis and validation show that the proposed method can provide guarantee conditions for network survivability effectively.