Federated Self-Supervised Learning (FSSL) is a promising paradigm for extracting robust representations from decentralized unlabeled data. However, its effectiveness is often hindered by non-IID data distributions and label scarcity, which cause model divergence and limit generalization. In this paper, we propose Federated Self-Supervised and Global-Personalized Collaborative Learning (FedGP), a novel framework designed to bridge the gap between global knowledge integration and local client adaptation. The core of FedGP is the Collaborative Knowledge Anchoring (CKA) mechanism, which utilizes adaptive regularization to anchor shared global knowledge while enabling personalized refinement on local data. By dynamically balancing collaborative risks and local empirical losses via learnable coefficients, FedGP ensures stable convergence in heterogeneous environments. Extensive evaluations on multiple benchmarks, including a real-world private Flora dataset, demonstrate that FedGP consistently outperforms state-of-the-art FSSL methods. Our results confirm that FedGP achieves high-quality representation learning with significantly reduced communication overhead and annotation dependency, providing a scalable solution for privacy-preserving decentralized systems.

The deep integration of Internet of Vehicles (IoV) and edge computing technologies brings new requirements for the Vehicular Edge Computing (VEC) information system security evaluation. Facing the two core problems of resource-constrained scenarios and dynamic security evaluation, the GFCIV-CGTOPSIS model for VEC information system dynamic security evaluation is proposed. In the model, subjective and objective evaluation indexes are considered and calculated separately to improve operability. An improved grey correlation F-statistics clustering and index validity combination index screening (GFCIV) method is proposed in order to improve the operational efficiency of the traditional grey F-statistics rough set (GFRS) index screening method. The CRITIC method is used to determine the subjective and objective comprehensive index set weights, and the grey TOPSIS method is used to realize the dynamic security evaluation. Experimental results demonstrate that the GFCIV-CGTOPSIS model, compared to the pre-improved GFRS-CGTOPSIS model, achieves reduced index tree redundancy and efficient dynamic security evaluation while exhibiting less information loss and lower evaluation result deviation due to index screening.

In the era of big data and IoT, personalized federated learning (pFL) addresses privacy challenges by keeping training on-device and transmitting only parameter updates. This approach allows each client to incorporate their unique data characteristics, effectively tailoring the global model to diverse user needs. However, as pFL leverages personalized updates that capture individual client experiences, it faces the inherent risk of exposing sensitive information during the aggregation process. Global federated methods like FedAvg further exacerbate this issue when the non-IID assumption is violated across clients, as they struggle to maintain performance across varied data distributions, often leading to suboptimal results. To tackle this, we propose fine-grained Privacy-aware Parameter Coaching method for personalized Federated learning(PPCFed). We propose a dynamic privacy matrix that quantifies layer-wise privacy leakage risks and adaptively reweights inter-client knowledge transfer. This matrix acts as a fine-grained privacy-aware controller, enabling clients to selectively assimilate insights from others without additional overhead while actively suppressing high-risk information flows. Our method balances privacy and performance, showing improved results in heterogeneous FL and pFL environments.

Quantifying the deviation of testing samples from known benign traffic in the form of novelty scores is essential for identifying new malicious traffic and detecting concept drifts of benign traffic. Most existing solutions generate novelty scores based on the outputs from the supervised neural network’s last layers. However, the performance of these supervised techniques is significantly compromised in the unbalanced dynamic interconnected specialized IoT network traffic, such as electronic power grid. To solve this challenge, we investigate an improved novelty score calculation approach. It employs the targeted distance loss to ensure that different known classes form class-specific, dense clusters within the embedding space outputted by the neural network. The minimum distance between the testing sample and the centers of each known class is utilized as the novelty score. Extensive comparison experiments on public datasets and real Electronic Power Grid traffic demonstrate that our approach outperforms existing supervised techniques. As the imbalance in known class distributions increases, our proposed method consistently achieves higher AUROC and AUPR than baselines, with a reduced FPR95. This study analyzes imbalanced known class distributions’ negative impact on novelty detection and provides practical improvement.

Due to the difficulty in predefining novel attack patterns and the scarcity of sufficient malicious training samples in specialized Internet of Things (IoT) scenarios, the focus of IoT intrusion detection researchers has shifted toward anomaly-based techniques like Autoencoder. These techniques detect attacks based on the degree of deviation and rely minimally on malicious samples. However, current machine learning (ML) and deep learning (DL) implementations lack explainability. Although some methods provide post-hoc interpretations, they are limited and partial, failing to reflect the entire decision-making process. To address this challenge, we investigate an explainable anomaly-based intrusion detection system (IDS) that translates the inference process of the Autoencoder into the high-fidelity allow-list rule library, thereby balancing the detection capability and interpretability. First, we assume that benign traffic follows a complex global distribution composed of several irrelevant local distributions. The clustering algorithm is performed in an extended feature space consisting of reconstruction loss and embeddings to decompose local distributions. Then, we deploy an approach based on Gradient Ascent to explore the boundary rules of each local distribution. The allow-list rule library that reflects Autoencoder’s inference process can be constructed by merging these boundary rules. Comprehensive evaluation experiments demonstrate that the extracted allow-list rule library accurately reproduces Autoencoder’s inference process and effectively detects IoT intrusions.

Based on the rapid development of edge computing and resource-constrained characteristics, new requirements for edge information system security evaluation are proposed. Oriented to the resource-constrained scenarios of edge information systems and the distribution characteristics of raw data for security evaluation, the adaptability of existing models is improved. The improved Principal Component Analysis (PCA-S method) based on Spearman’s Coefficient is proposed for index screening. In order to further improve the screening effect and reduce the resource consumption of the security evaluation model, the PCA-S method and the Distinction Degree Screening method are combined by taking the intersection, and the PCA-SDC combination screening method is proposed. Through the PCA-SDC index combination screening, the Coefficient of Variation weighting, and the Fuzzy Comprehensive Evaluation, the PSDC-CVF edge information system security evaluation model is finally formed. In order to assess the effect of index screening and security evaluation, three indicators, namely, the improved Average Quantity of Information Change Degree, the Average Information Contribution Change Degree, and the Fuzzy Evaluation Deviation Degree, are proposed. Through experiments, the improved PCA-S method and the combination screening PCA-SDC method are sequentially proved to be well adapted and effective in the index screening process of edge information system security evaluation. It is also verified that the PSDC-CVF model reduces the resource consumption compared with the traditional model and better balances the model energy consumption and performance.

The frequent attacks show that no information system is absolutely safe and the security capabilities are relative. Security resilience becomes a complementary priority for improving information systems’ continuous service and security capabilities in the face of attacks, such as unknown vulnerabilities and backdoors. Endogenous security defense technology has become an important research aspect to improve the security resilience of information systems. However, there are some limitations in the research of the information system security resilience evaluation model, such as lacking indexes to characterize the system security resilience under an attack environment. In this article, a security resilience enhancement strategy based on dynamic defense is constructed to improve the security performance of the system through IP port hopping and attack surface conversion. For the adversarial behaviors of attackers and defenders, we propose a security resilience metric framework based on the evolution of attack and defense scenarios, which is evaluated using a resilient security evaluation model based on the fuzzy Choquet integral. In the model, the weights of evaluation indicators are calculated based on the decision-making trial and evaluation laboratory method. The 2-addable fuzzy measures of each indicator are calculated secondarily. Then the security performance of the system is calculated using the fuzzy Choquet integral. Absorptive capacity, adaptive capacity, and resilience factor are proposed to better supervise the metric framework’s validity. Finally, four groups of control cases were created by building the Web service system after the endogenous security transformation as the experimental simulation scenario. Experimental simulation results show the superiority of the proposed metric model.

In view of the difficulty in determining reasonably evaluation indicator system in the information security certification and accreditation work, an information security evaluation framework supporting the adaptation of measurement models is proposed. And the mapping-based information security evaluation indicator construction rule and a measurement model library are established. The optimal measurement model is adapted according to head-to-tail consistency and standard deviation index. Then, the evaluation model relies on the information security index feedback algorithm based on probability iteration to adjust the evaluation indicator system for more reasonableness. This paper provides a model reference for information security certification.