With advancements in machine learning (ML) technology and the deployment of large ML-as-a-Service (MLaaS) clouds, accurately understanding request behaviors in an MLaaS cloud platform is paramount for resource scheduling and optimization. This paper sheds light on the correlation of request arrivals in a representative and dynamic MLaaS workload – Alibaba PAI (an ML platform for artificial intelligence). For requests in the PAI workloads at the job, task, instance, and machine levels, our burstiness diagnosis reveals that the request arrival processes at all levels are significantly bursty. Additionally, our Gaussianity test indicates that the bursty activities in PAI consistently appear to be non-Gaussian. Our findings show that there exists a certain degree of correlation between request arrivals at each level over long-term time scales. Moreover, we reveal the self-similar nature of request activities in the various-level wild MLaaS workloads on Alibaba PAI through visual evidence, the auto-correlation structure of the aggregated process of request sequences, and Hurst parameter estimates. Furthermore, we implement a versatile workload synthetic model to synthesize request series based on the inputs measured from the PAI trace. Experimental results demonstrate that our model outperforms typical self-similar workload models, and can improve accuracy by up to 99% compared to them.

To address the challenges of data insufficiency, model complexity, privacy protection, and zero-trust continuous authentication in intelligent connected vehicle driver identification, this paper proposes a Zero-Trust Driver Continuous Authentication System that integrates Federated Learning with Conditional Transformer GAN (FCG). FCG employs a structurally concise Memory-Augmented Autoencoder to achieve efficient feature learning, and proposes a Transformer-based GAN model for data augmentation. It innovatively utilizes existing sensor data as conditional input to generate cross-physical quantity data, which both alleviates the data scarcity problem and preserves the natural correlations between different data types. For privacy protection, FCG combines the Memory-Augmented Autoencoder with a federated learning framework, achieving distributed collaborative training where 'models move instead of data,' while simultaneously meeting the requirements of continuous verification under the zero-trust security model. Extensive experiments on authoritative datasets demonstrate that FCG significantly outperforms existing methods in terms of false alarm rate and authentication accuracy. The research results of this paper are expected to promote the development and application of identity authentication technology for intelligent connected vehicles.

Federated learning enables collaborative model training across multiple parties while preserving data privacy. Traditional frameworks assume the server only aggregates models, making them unsuitable for leader-follower scenarios. To address this, we propose a leader-follower federated learning framework, where the server, with limited data but strong computational capacity, leverages client data for additional training. We introduce Federated Asymmetric Distillation (FedAD) to accommodate distinct server and client objectives: the server employs a weak-to-strong distillation strategy to correct inaccuracies from client models, while clients use Not-True Distillation to mitigate catastrophic forgetting. Comparative and ablation experiments show that, after federated training, the Res110-20 model improves accuracy by 14.55%, 7.15%, and 4.80% on CIFAR100, CIFAR10, and MNIST, respectively.

Federated learning (FL) enables collaborative training of global models with local data, preserving privacy. However, its parameter update process exposes it to Byzantine attacks by malicious clients, degrading model performance. We propose FedPMR: first, PCA reduces dimensionality of client-uploaded parameters to lower clustering costs; second, a dual detection module identifies anomalies via global deviation and local structural similarity; finally, a robust aggregation module integrates detection results with historical scores to determine client weights. Experiments show FedPMR effectively suppresses malicious clients, maintains accuracy, reduces overhead, and exhibits strong robustness and adaptability.

Hierarchical federated learning (HFL) has attracted significant attention for its communication efficiency and cost-effectiveness. However, its distributed nature makes it vulnerable to malicious attacks, particularly in large-scale, zero-trust edge networks where monitoring nodes is challenging. Existing defenses for traditional two-layer federated learning are insufficient to address the unique cross-layer collusion attacks possible in HFL. To bridge this gap, we introduce HFLMLD, a robust HFL framework with Multiple Layer Defenses. HFLMLD employs a two-stage hierarchical defense strategy to enhance system resilience. At the edge layer, it combines dimensionality reduction-based detection with a dynamic suspicion score mechanism to identify and neutralize malicious clients. At the cloud layer, a normalized weighted aggregation algorithm is employed to counter threats from compromised edge servers. Our extensive experiments on benchmark datasets show that HFLMLD effectively secures HFL systems against complex, multi-faceted attacks while maintaining high model performance.

In recent years, Industrial Control Systems (ICSs) have faced increasing vulnerability to cyber attacks due to their integration with the Internet. Despite efforts to enhance cybersecurity, reconnaissance attacks remain a significant threat, prompting the need for innovative defensive strategies. This paper introduces a novel approach to strengthen the defensive capabilities of ICS networks against reconnaissance attacks using machine learning-driven cyber deception techniques. Leveraging Conditional Generative Adversarial Networks (CGANs), the proposed framework dynamically generates defensive network topologies to network shuffling and implement deception strategies, prioritizing system availability. Extensive simulations demonstrate the superior efficacy of the proposed framework in enhancing cybersecurity while minimizing computational overhead. By effectively mitigating reconnaissance attacks, this solution reinforces the resilience of ICS networks, safeguarding critical industrial infrastructure from evolving cyber threats. These findings underscore the significance of adopting machine learning-based cyber deception as a pragmatic security measure for protecting ICS networks in real-world industrial contexts.

Federated Learning (FL) has drawn much attention for distributed system over the Internet of Things (IoT), since it enables collaborative machine learning on heterogeneous devices while resolves concerns about privacy leakage. Due to the catastrophic forgetting (CF) phenomenon of optimization methods, existing FL approaches are restricted to single task learning and typically assume that data from all nodes are simultaneously available during training. However, in practical IoT scenarios, the data preparation from nodes may be asynchronous, and different tasks require incremental training. To address the issues, we propose a continual FL (CFL) framework with parameter-level importance aggregation (PI-Fed), which supports collaborative task-incremental learning with privacy preservation. Specifically, PI-Fed evaluates the importance of each parameter in the global model to all history tasks, which is computed locally and aggregated at the center server. Then the server performs soft-masking on the averaged gradient collected from local clients based on the parameter importance. By minimizing the change on important parameters, PI-Fed effectively overcomes CF and also achieves high efficiency without experience replay. Extensive experiments on 4 benchmarks with at most 20 sequential tasks demonstrate that our proposed PI-Fed significantly outperforms traditional FL baselines (FedAvg, FedNova, and SCAFFOLD).

Term polarity co-extraction is an aspect-based sentiment analysis task, which has been widely used in the fields of user opinions extraction. It consists of two subtasks: aspect term extraction and aspect sentiment classification. Most existing studies solve aforesaid subtasks as independent tasks or simply unify the two subtasks without making full use of the relationship between tasks to mine the interaction of text information, which leads to low performance for practical applications. Meanwhile, the learning framework for these studies has a label drift phenomenon (LDP) in the process of predictive learning, increasing the learning error rate. To address the above problems, this study unifies subtasks and proposes a Unified framework based on the information interaction mechanism framework, called IIM. Specifically, we design an Information Interaction Channel (IIC) to construct closer semantic features to extract preliminary term-polarity unified labels from the perspective of basic semantics. For label inconsistency between aspect terms, a Position-aware Module (SAM) is proposed to alleviate the Label Drift Phenomenon (LDP). Moreover, we introduce a syntax-attention graph neural network (Syn-AttGCN) to model the syntactic structure of text and strengthen the emotional connection between aspect terms. The experimental results show that IIM outperforms most baselines. Meanwhile, the SAM module has a certain slowing effect on LDP.

As big data, medical digitalization, and wearable devices continue to evolve, these technologies are driving the advancement of clinical medicine, genomics, and wearable health while also posing a risk of privacy breaches. For the disclosure of medical privacy issues, some methods and systems combining differential privacy protection applied in the clinical, genomic and wearable fields have been summarized, analyzed and compared. Finally, through the investigation and research, the assumptions and prospects for future research hotspots are given.

The business relationships between Autonomous systems (ASs) are crucial to understanding the internet structure, performance, evolution, and so on. However, the business relationships between ASs are often confidential and can only be captured by inference algorithms. This paper proposes a new algorithm to infer the ASs relationship based on the transmission capacity. A new metric for ASs node transmission capability based on path behavior is defined. Meanwhile, a big data processing technique based on Map-Reduce to deduce the ASs relationship. Besides, to analyze the accuracy of the algorithm, we compare its accuracy performance with those of existing algorithms. Through simulation experimental, the proposed scheme achieves substantial performance improvements in term of consistency and effectiveness. Also, the complexity of the proposed scheme is found to be reasonably low.

In large-scale network service systems, the phenomenon of instantaneous gathering of a large number of users can cause system abnormality, whenever the load imposed by the user behaviors does not match the system load. This paper proposes a behavior reconstruction model for large-scale network service systems integrated with Petri net reconstruction methodology, for the purpose of achieving load balancing in the system under increasing number of users. Based on the features of the user interaction behavior sequence, the behavioral load balancing model defines a user behavior membership function. Then, a random fuzzy Petri net with delay is presented to control the user behavior reconstruction. Experiments conducted by considering various changes in the number of user behaviors and their distribution in unit time demonstrate that the proposed methodology can effectively trigger the reconstructed model to balance the system load when the system load exceeds the defined warning point.

Multipath inter-domain routing can improve the reliability, robustness and path diversity of the Internet. This paper proposes one User-customizing oriented Multi-path Inter-domain Routing protocol called as UMIR. Its basic idea is as follows: based on user routing requirements, it selects some nodes from the feasible BGP paths and requests their path-lets information, and then it builds a local topology to calculate user routes. The experimental analysis shows that this UMIR protocol can achieve rich candidate and high-quality paths.

In the grid environment, the resource providers should maintain the ultimate authority over their resources, including access authorization and grain scale control. Apparently, the traditional RBAC model will be inappropriate in multi-autonomous domains environment. Focusing on the autonomous, heterogeneous and dynamic features of grid computing, we propose the concept of "Resource Role". We also initiate the dual-role based access control (DRBAC) framework, where the resource role permission mapping is defined in resource domain and user role resource role mapping is negotiated by user domain and resource domain. This framework is simple and works better than the traditional RBAC in multi domains environment