Federated Learning (FL), as a distributed training framework, has demonstrated significant potential in preserving data privacy. However, client heterogeneity in data volume, computational resources, and communication conditions causes two critical issues: inefficient global synchronization due to stragglers, and excessive communication overhead from repeated model transmissions. To address these challenges, this paper proposes an Adaptive Federated Distillation (AFD) framework, which enhances training efficiency by dynamic local training optimization and balanced communication load. AFD dynamically adjusts the number of local training epochs based on client-specific data volume, communication conditions, and historical training time. Simultaneously, it employs federated distillation to transmit lightweight logits instead of full model parameters, thereby reducing communication costs and enabling model heterogeneity. Experimental results on the MNIST and CIFAR-10 datasets demonstrate that, compared to FedAvg, AFD reduces communication costs by 3-4 orders of magnitude when CNN or ResNet models are used. Additionally, AFD reduces client energy variance by 96.5% and training time disparity by 88.7% while maintaining model accuracy. Crucially, its model-agnostic design supports heterogeneous architectures, ensuring fairness across diverse edge devices.

Flexible manufacturing systems (FMS), empowered by the Industrial Internet of Things (IIoT), have become a cornerstone of Industry 6.0 by enabling dynamic production adaptation, real-time equipment monitoring, and intelligent scheduling. As these systems increasingly incorporate large language models (LLMs) to support functions such as knowledge querying, decision assistance, and predictive maintenance, ensuring the safety and reliability of human-LLM conversations has become a pressing concern. Specifically, LLMs may generate toxic, biased, or privacy-violating outputs when interacting with sensitive IIoT data and production logic, potentially compromising operational safety. To address this challenge, we propose AugLLMSen, an automated toxicity evaluation framework tailored to the IIoT-driven FMS context. AugLLMSen integrates a question automatic expansion mechanism (Q-Judge) and an output toxicity evaluation model (O-Judge) into a closed-loop pipeline, enabling large-scale assessment of LLM safety across diverse industrial scenarios. Experimental results on open- and closed-source LLMs demonstrate the effectiveness and accuracy of our approach in identifying toxic responses and guiding safe deployment of LLMs in flexible manufacturing environments.

With the volume of data increasing in the Internet of Things, a new business mode, where data owners share their own data to others for rewards, has emerged. Therefore, how to motivate data owners to participate in the data trading process is the main challenge. So far, lots of works focus on the motivation mechanism designing and ensure a fair distribution of profits among data owners. However, some security and privacy issues are still not well solved and the data owners are still unwilling to participate in the process. Especially, when a data provider claims rewards with its real identity for the shared data, the linkage between its real identity and the shared data will expose the participator's private information included in the shared data, such as location information. To protect user's privacy in the scenario, a privacy and efficiency-oriented data sharing mechanism for IoTs is proposed in this paper. We first propose a blockchain-based data sharing framework in which the behavior of all participants will be supervised. Then, in order to hide the real identities of data providers during the data sharing process, an anonymous certificate-based data sharing policy is proposed. At last, two novel non-interactive zero-knowledge proofs are designed to hide the identities of qualified data providers while claiming rewards to the system. Through security analysis and performance evaluation, the feasibility and effectiveness of the data sharing scheme are illustrated.

Nowadays, the quality of cloud services offered by different service providers varies greatly. Reputation mechanism, as a better service evaluation method, can help standardize and regulate the cloud service market. However, existing reputation systems either rely on a trusted third party with security and privacy issues, or lack reliable evaluation. To address the above issues, we propose a blockchain based privacy-preserving dynamic reputation mechanism for cloud service and a reputation management smart contract (RM) is designed to implement trusted reputation computation. The reputation integrates conformance trust in the subjective view and recommendation trust in the objective view together to provide a comprehensive evaluation. Besides, a miner selection algorithm is designed to prevent miners from launching a collusion attack. Moreover, the Paillier homomorphic encryption algorithm (PHE) is introduced to encrypt the sensitive data of customers, ensuring the security of data stored on the blockchain. Experiment results reveal that the proposed model is feasible and the performance of encryption is acceptable.

As the intelligentization of Industrial Internet of Things (IIoT) broke the relatively closed and credible industrial environment, IIoT faces increasingly serious security problems. The commonly used vulnerability discovery method is protocol reverse engineering. However, it is difficult to analyze IIoT protocols with existing protocol reverse engineering approaches, as they influence the normal operation or have spare sample data. In this paper, a sparse protocol parsing method for IIoT protocols is proposed. The parsing method expands the samples of the captured IIoT protocol message sequences using a genetic algorithm (GA), which designs its fitness function based on the protocol response data to select high-quality samples. By combining the GA with the hidden Markov model (HMM) with lower algorithm complexity, a hybrid parsing model is constructed to improve accuracy in a gradual evolution way. Through comparison experiments on various IIoT protocols, our HMM hybrid model has better performance than RNN hybrid models under sparse samples.

Renewable energy powered road side units(RSUs) in Internet of Vehicles(IoVs) are a desirable green alternative choice compared to the traditional electric grid, because it not only extends the service range of IoVs but also saves the cost on energy. However, the energy on RSUs is limited and usually influenced by communication policies. Therefore, the communication scheduling policy on RSUs plays an important role on the persistence of network, which is still an open topic to be solved. In this paper, we focus on the communication scheduling problem on RSUs, and propose an LSTM-based communication scheduling algorithm for RSUs in IoVs. The scheduling algorithm is composed of three parts - deep learning clustering, LSTM-based traffic prediction, and a vehicle access scheduling algorithm. At last, we conduct an extensive simulation, and the simulation results indicate that our algorithm can achieve a better performance than the no scheduling mechanism.

Advertisements(Ads) dissemination in Vehicular Ad Hoc Networks (VANETs) is a promising application. It not only provides benefits to customers but also brings commercial profits to merchants. However, little previous work has focused on the ads dissemination problem to maximize the overall profit. In this paper, we target the ads dissemination problem and design two algorithms for ads delivery. We first propose a Spread Utility Based Delivery Algorithm based on the knowledge derived from the empirical study of vehicle traces. Based on the first algorithm, an Optimized Ads Delivery Algorithm is designed, which considers consumers' interests and their mobilities. At last, we conduct extensive simulation, and the simulation results indicate that our algorithms can achieve a higher delivery rate and effective rate than current algorithms in most cases. Additionally, when the consumer interest is considered, our algorithm can achieve a tradeoff between the delivery rate and visiting probability.