Mobile Crowd-Sensing (MCS) has emerged as a promising solution for large-scale data collection in Internet of Things (IoT) scenarios. However, the malicious behavior of smart terminals, such as providing corrupted and falsified data or deliberately spreading false data, poses a significant threat to the credibility of MCS services. At present the mainstream trust evaluation schemes evaluate its trust value through the accumulation of terminal interaction experience, but the inherent defects of these schemes will make MCS service suffer serious trust-decaying destruction. And the current trust computing methods do not yet take into account the balance between the quality of service completion and the crowd-sensing collaboration experience of the terminal. To solve these problems, this paper proposes a novel general grey buffer trust evaluation framework (TrustGrey), which is used to evaluate the trust relationship of smart terminals. Specially, we construct the trust value calculation model of grey trust state for smart terminal to make up for the inherent defects of normal trust value calculation model. The grey buffer of sudden drop is designed in the trust value calculation model to avoid trust-decaying destruction. And we design a quick recovery mechanism of grey buffer to avoid detecting false alarm caused by the sudden drop of trust, as well as avoiding the loss of multiple damage of intelligent malicious terminal by an irreversible black value reduction mechanism. Then, we design a supply-demand equilibrium based dynamic recruitment mechanism to dynamically coordinate the recruitment process of service requests by comprehensively considering the importance level of service requests and the credibility of terminals, so as to balance the experience of service originators and completion parties. Experiments on real word datasets highlight the advantages of our proposed framework TrustGrey. And, the experiments also show that TrustGrey has considerable versatility.

Class-incremental learning (CIL) aims to enable a learning system to continuously learn new information. Although pre-trained models (PTMs) exhibit strong performance in CIL, the lack of data from previously learned classes results in class imbalance and catastrophic forgetting during the updating process. Consequently, integrating new categories while retaining previous knowledge remains a significant challenge. In this paper, we introduce the Adapt and Feature Translation (AFT) method for PTM-based CIL, designed to enhance the stability-plasticity balance. The adapter module is employed to adapt the model for specific downstream tasks, while keeping all backbone parameters fixed during subsequent updates to maintain the stability of the feature extraction. Additionally, actual features of new classes, combined with synthesized features of old classes via translation, are jointly used to retrain the classifier, ensuring a balanced classifier. Experimental results on multiple datasets validate the effectiveness of our method, achieving a new state-of-the-art performance.

The field of medical diagnostics is undergoing an AI-driven transformation from traditional centralized laboratory testing to instantaneous testing. Federated Learning (FL) has emerged as a pivotal technique in medical image feature engineering. However, medical FL faces challenges such as vulnerability to data poisoning and inefficiency in handling dynamic class distribution changes in dynamic and noisy environments. To address these issues, we propose a dynamic FL framework called ADFL-DSV. The framework introduces a class-specific Shapley computation method for dynamically evaluating the contributions of clients during training. To enhance robustness in dynamic environments, we also develop an adaptive weighting mechanism inspired by dynamic Shapley values. Additionally, we accelerate the sampling process using client-importance sampling, optimizing the global update variance based on client weights and reducing communication costs. We validate our approach on several real datasets, outperforming benchmark approaches by$\mathbf{7 \%}$to$\mathbf{1 4 \%}$in terms of performance.

Dynamic trust evaluation in online social networks is of great significance for applications such as recommendation systems, financial risk control, and e-commerce, but it remains a challenging research problem. Most existing methods are based on static network topologies or simple temporal modeling, making it difficult to effectively capture the dynamic evolution laws of trust relationships. To address this issue, this paper proposes a dynamic graph neural network-based trust evaluation model, TeGCN, for predicting dynamic trust relationships in social networks. TeGCN consists of a spatial aggregation unit, a gated temporal evolution unit, and a joint prediction unit. In the spatial aggregation unit, node spatial feature representations are generated through linear transformation after concatenating features of neighbor nodes and forward/backward edge attribute information via average aggregation. The gated temporal evolution unit employs a Gated Recurrent Unit (GRU) to model the temporal evolution of trust relationships and introduces a dynamic weight adjustment mechanism to enhance feature extraction at key time steps. Finally, the joint prediction unit integrates spatio-temporal features, performs nonlinear mapping through a multi-layer perceptron (MLP), and outputs the probability distribution of node trust levels to achieve end-to-end dynamic trust prediction. Experiments based on two real dynamic trust network datasets, Bitcoin-Alpha and Bitcoin-OTC, verify the effectiveness of TeGCN. The results show that TeGCN significantly outperforms existing baseline models in key metrics such as F1-score, AUC, and ACC, especially in highly dynamic scenarios.

As a cutting-edge and visionary computing mode, edge computing is expected to significantly improve the computing performance of mobile devices (MDs) and improve user experience. This improvement is due to the shift of compute-intensive tasks from MDs with limited computing power to the more well-resourced cloud. However, the task unloading process also faces the limitation of limited communication bandwidth and server resource usage. The task of minimizing the total overhead of the system is especially difficult when the application needs to be completed in a certain amount of time. To overcome these obstacles, our research proposes a novel task offloading and resource allocation strategy based on graph neural networks (GNN). Our strategy can assist in decision-making on computing task offloading, controlling CPU clock frequency, and allocating communication capabilities. To verify the validity of our approach, we conducted a series of simulation experiments in different scale edge computing scenarios. The experimental results show that the task unloading optimization algorithm based on graph neural network can guarantee the quality of service (QoS) and manage the whole system resource cost reasonably. Therefore, our study provides a substantial basis for further exploration of GCN’s potential in the field of edge computing task offloading decision making.

Edge computing has emerged as a popular paradigm for handling heterogeneous tasks to improve computing capacity and promote quality of service (QoS). Multi-access edge computing can balance overall benefits while increasing the efficiency of edge servers. However, the design of computation offloading is challenging. This study provides a Graph Attention Network Reinforcement Learning (GATRL) framework to address the offloading challenge, which models devices and edge servers as digraphs and the state policy by graph properties to address the problem. The GATRL uses supervised learning and combines the GAT and DQN. We propose a method to identify the approximate solution to achieve fast convergence based on the data generated by Gaussian distribution relied on actual parameters. Compared to other methods, numerical results show that the proposed GATRL algorithm can achieve near-optimal performance and significantly reduce the off-load time. In addition, graph neural networks can adapt to more variable network environments, thus rapidly changing the model structure and giving offloading solutions.

Federated learning allows participants to share models (gradients) rather than raw data to collaboratively train a global model, enhancing participants’ privacy protection but making the global model more vulnerable to poisoning attacks. Poisoning attacks not only lead to the degradation of model performance but also cause a security risk. A mainstream defense strategy against poisoning attacks is that the server identifies malicious models by analyzing the models uploaded by participants. However, the attackers can use the models uploaded by the participants to recover their privacy data, leading to a privacy disclosure. Therefore, participants need to encrypt or perturb the models before uploading them so that all individuals, including the server, cannot access the model, which poses a great challenge for the defense against poisoning attacks. Moreover, many existing defense strategies against poisoning attacks only work well when the data distribution is non-independently and identically distributed. To address these issues, we propose a novel defense strategy for poisoning attacks of federated learning called FL-PTD, which can judge whether the global model is subjected to poisoning attacks without accessing the participant’s local model. Besides, Our method incorporates a trust evaluation mechanism, which computes reputation scores based on the historical behavior of participants to identify malicious participants. Finally, we verify our proposed method on several real world datasets. The experimental results show that our method can effectively defend against poisoning attacks and accurately identify attackers without compromising the model’s performance.

In order to satisfy the trust evaluation of efficient cooperative scheduling of computing power in Computing Power Network, we propose a trust evaluation management system with adaptive detection and an efficient lightweight trust evaluation algorithm. In our work, the multi-attribute trust evaluation data combined with the active detection and global trust database are used as samples to train the BP neural network. And the structure and weight coefficient of the neural network are optimized by the improved particle swarm optimization algorithm, so as to reduce the size of neural network and improve its performance. The trust evaluation model in this study effectively improves the detection ratio of malicious state nodes and reduces the detection time.

With the growth of IoT and latency-sensitive applications (e.g., virtual reality, autonomous driving), massive amounts of data are generated at the network and the edge. Such proliferation drives the development of dispersed computing as a promising complementary paradigm to cloud computing and edge computing. Dispersed computing can leverage in-network computing resources to provide lower latency guarantees and more reliable computing power support than the cloud computing. On the other hand, dispersed computing shows excellent performance in highly dynamic and heterogeneous environments. This paper introduces the concepts of dispersed computing and demonstrates the potential of dispersed computing in terms of providing low-latency services and adapting to highly dynamic and heterogeneous environments through several cases. To better grasp the current state of research in dispersed computing, several major research directions and advances are also presented in the hope of attracting the attention of the community and inspiring more researches to promote the implementation of dispersed computing.