Future Generation Computer Systems-The International Journal of Escience最新文献

{"title":"FedDSKI: Improving server-side model via dual-stage knowledge isolation in personalized federated learning","authors":"Xiaorui He ,&nbsp;Jinjia Peng ,&nbsp;Zhen Wang ,&nbsp;Hui Li ,&nbsp;Huibing Wang","doi":"10.1016/j.future.2025.108061","DOIUrl":"10.1016/j.future.2025.108061","url":null,"abstract":"<div><div>In statistically heterogeneous scenarios, traditional federated learning is unable to adapt to all data distributions. Consequently, personalized federated learning (PFL) is developed to customize models for each client to ensure adaptation for local data distribution. Recent methods in PFL focus on obtaining common knowledge to enhance local training. However, the issue of declining generalization performance in the server model is often overlooked. To tackle this issue, this paper proposes a personalized federated dual-stage knowledge isolation method (FedDSKI) to reduce knowledge degradation during knowledge transfer. FedDSKI mainly includes two components: Domain-Centered Regularization (DCR) and Category Knowledge Recovery (CKR). DCR regulates local training by infusing global information into the clients. This global information provides a consistent optimization direction for the clients in the representation space, thereby eliminating client bias during the common knowledge learning stage. CKR emphasizes extracting local knowledge in a category perspective, utilizing prototypes to refine category information. During the personalized knowledge learning stage, CKR dynamically captures the unique knowledge of each class in the local dataset, guided by real-time updates of local historical prototypes. To validate the effectiveness of FedDSKI, extensive experiments were conducted on three widely used computer vision (CV) datasets. The results demonstrate that the proposed method outperforms the baselines.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108061"},"PeriodicalIF":6.2,"publicationDate":"2025-08-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144830712","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CANDICE: An explainable and intelligent framework for network intrusion detection","authors":"Shuhua Li ,&nbsp;Ruiying Du ,&nbsp;Jing Chen ,&nbsp;Kun He ,&nbsp;Cong Wu ,&nbsp;Yebo Feng","doi":"10.1016/j.future.2025.108059","DOIUrl":"10.1016/j.future.2025.108059","url":null,"abstract":"<div><div>In recent years, Deep Learning-based Network Intrusion Detection System (DL-NIDS) have demonstrated remarkable performance in detecting cyberattacks in network traffic. However, the lack of explainability for DL-NIDSs prevents end-users from trusting and understanding the detection results, thereby limiting their applications in practice. Although several approaches have been proposed to explain DL-NIDS, they run the risk of providing unfaithful explanations. In addition, existing methods merely output a set of important features as explanation, which is insufficient for end-users to thoroughly understand the attack. In this paper, we propose CANDICE, an explainable and intelligent framework for detecting and explaining intrusions in network traffic. Differing from existing works, CANDICE is highlighted by: (i) providing faithful explanation by disentangling the traffic representations and generating counterfactual explanations, and (ii) offering end-users a comprehensive view of the attack by generating an intrusion profile based on the explanation. We conduct experiments on four representative traffic datasets to evaluate the effectiveness of CANDICE. The results demonstrate that CANDICE surpasses existing methods in terms of explanation fidelity, sparsity, stability, and efficiency, while achieving high accuracy of above 96.10% in detecting intrusions.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108059"},"PeriodicalIF":6.2,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144827641","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cheesemap: A high-performance point-indexing data structure for neighbor search in LiDAR data","authors":"Ruben Laso ,&nbsp;Miguel Yermo","doi":"10.1016/j.future.2025.108060","DOIUrl":"10.1016/j.future.2025.108060","url":null,"abstract":"<div><div>Point-cloud data, as the representation of three-dimensional spatial information, is a fundamental piece of information in various domains where indexing and querying these point clouds efficiently is crucial for tasks such as object recognition, autonomous navigation, and environmental modeling. In this paper, we present a novel data structure, <span>cheesemap</span>, designed for fast neighbor search in 3D LiDAR point clouds. Points are indexed using a grid of voxels, which can be organized in three different ways, originating three flavors of the <span>cheesemap</span>: dense, sparse, and mixed. The lookup of the voxels is theoretically ensured to be performed in constant or amortized constant time, speeding up the search for neighboring points. Experimental results show that <span>cheesemap</span> can outperform, in terms of performance and memory footprint, other state-of-the-art data structures both in region-based and <span><math><mi>k</mi></math></span>-NN queries throughout different types of point clouds, particularly for Airborne Laser Scanning (ALS) point clouds.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108060"},"PeriodicalIF":6.2,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144809900","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optimized node placement and dynamic session node selection for permissioned blockchain in industrial IoT","authors":"Saurav Gupta,&nbsp;Sukumar Nandi","doi":"10.1016/j.future.2025.108057","DOIUrl":"10.1016/j.future.2025.108057","url":null,"abstract":"<div><div>The Industrial Internet of Things (IIoT) increasingly leverages permissioned blockchains to address inherent security and efficiency challenges posed by centralized data architectures. While advancements have improved consensus mechanisms and transaction rates, optimally deploying these blockchains in resource-constrained and expansive IIoT networks remains a significant hurdle. One potential avenue for enhancing efficiency is by reducing the number of full blockchain nodes operating within the system. However, if not strategically managed, such a reduction can compromise network-wide device serviceability and undermine system security. Furthermore, traditional scaling solutions like sharding, while aiming for efficiency, introduce their own complexities and may not be ideal for all IIoT scenarios requiring robust, uniform security. This paper proposes a novel framework to optimize blockchain performance in IETF-based IIoT systems by enhancing operational efficiency while rigorously maintaining security. A cornerstone of the approach is an optimal full-node placement strategy, uniquely formulated as a weighted K-uniform hypergraph vertex cover problem. The strategy directly addresses the challenge of node reduction by minimizing the required blockchain nodes while guaranteeing K-hop service availability for all network devices, offering an alternative to common off-chain or sharding methods. Additionally, the paper introduces a dynamic and unpredictable session node selection algorithm for consensus, which bolsters security and scalability by randomizing verifier participation in each block session. Collectively, the framework is designed to improve scalability and throughput while carefully considering decentralization, security assurances, and communication overheads. Experimental validation demonstrates that the solutions significantly reduce communication overhead and storage requirements, thereby enhancing blockchain viability for demanding IIoT applications without compromising data immutability or system security.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108057"},"PeriodicalIF":6.2,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144809901","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"IORT-DAG: A real-time DAG-based blockchain with implicit ordering","authors":"Guoqiong Liao ,&nbsp;Hao Ding ,&nbsp;Yinxiang Lei","doi":"10.1016/j.future.2025.108041","DOIUrl":"10.1016/j.future.2025.108041","url":null,"abstract":"<div><div>Directed Acyclic Graph (DAG)-based blockchains represent a paradigm shift from traditional blockchains, significantly improving throughput performance by concurrently committing blocks. Some DAG-based blockchain applications have real-time requirements, where transactions must be committed before specific deadlines. However, the ordering phase remains a bottleneck in DAG-based blockchains. DAG-based blockchains typically adopt an order-execute (OE) framework to resolve transaction conflicts. This explicit ordering approach requires dedicated time to reconcile transaction conflicts, leading to reduced throughput. For time-sensitive transactions, this overhead becomes critical as no transactions are processed during this period. Therefore, this paper proposes IORT-DAG, a DAG-based blockchain that employs an implicit ordering mechanism for transaction processing. Specifically, IORT-DAG adopts a merged execution and ordering concurrency control (MEOCC) to execute transactions concurrently. To ensure conflict serializability, we introduce the concept of a transaction anchor table, which stores temporary transaction states. Upon execution, transaction results are written into this table. Transactions failing to meet serializability conditions are restarted based on temporary states in the anchor table, while those satisfying serializability are committed according to their deadlines. To reduce tail latency, we propose a transaction position table that records transaction read-write sets after execution. This table enables the early commitment of execution completed transactions whose dependencies have been fulfilled, effectively reducing latency and enhancing throughput. Extensive experimental results demonstrate that IORT-DAG achieves robust performance across varying conflict rates and deadline constraints, effectively meeting real-time transaction requirements.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108041"},"PeriodicalIF":6.2,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144779403","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Trust-aware caching-constrained tasks offloading in multi-access edge computing","authors":"Xinyuan Zhu ,&nbsp;Fei Hao ,&nbsp;Ming Lei ,&nbsp;Aziz Nasridinov ,&nbsp;Jiaxing Shang ,&nbsp;Zhengxin Yu ,&nbsp;Longjiang Guo","doi":"10.1016/j.future.2025.108033","DOIUrl":"10.1016/j.future.2025.108033","url":null,"abstract":"<div><div>Multi-access edge computing (MEC) networks face significant challenges in managing congestion and safeguarding personal privacy data on a massive scale. Integrating trust awareness into MEC networks presents an opportunity to enhance security and privacy by correlating human relationships with connected devices. Moreover, leveraging trust-aware task caching and offloading holds promise in mitigating latency and reducing energy consumption. Despite existing research efforts to address these challenges, they often overlook either trust awareness or caching optimization in task offloading, potentially compromising security or leading to task failures. To address this gap, this paper proposes a novel approach: a trust-aware task offloading strategy with cache constraints (TCTO) in MEC networks, which considers social relationships, task offloading, and caching. Drawing on the characteristics of bipartite graphs and bipartite perfect matching, we develop a trust-aware caching-constrained task offloading algorithm based on bipartite graphs. This algorithm aims to select task offloading strategies that minimize delay, energy consumption in task transmission and execution, while maximizing security among devices in MEC networks. Extensive simulations demonstrate that our proposed method has a better performance than other task offloading strategies for reducing delay and energy consumption in the process of task transmission and execution. Compared with the other baselines, the overhead of our proposed method is reduced <span><math><mrow><mn>55</mn><mo>.</mo><mn>65</mn><mtext>%</mtext><mo>∼</mo><mn>96</mn><mo>.</mo><mn>20</mn><mtext>%</mtext></mrow></math></span> compared with other baselines.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108033"},"PeriodicalIF":6.2,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144773128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optimizing F2FS performance with the inter-zone parallelism in small-zone ZNS SSDs","authors":"Linbo Long ,&nbsp;Xinrui Dong ,&nbsp;Ting Wu ,&nbsp;Jingcheng Shen ,&nbsp;Kan Zhong","doi":"10.1016/j.future.2025.108044","DOIUrl":"10.1016/j.future.2025.108044","url":null,"abstract":"<div><div>ZNS SSDs are usually equipped with small zones to improve space utilization. However, a small zone is mapped to a limited number of chips, which results in a limited parallelism of the chips. Moreover, F2FS ignores the utilization of the inter-zone parallelism, resulting in an inability to fully utilize the chip parallelism of a ZNS SSD. On the one hand, inter-zone parallelism is rarely utilized when the request size of workloads is smaller than that of a zone. On the other hand, the inter-zone parallelism utilization is further limited by the inter-zone interference.</div><div>To address these issues, this paper proposes a full-parallelism zone allocation and writing strategy for F2FS, termed <em>FPZone</em>. The core idea is to allow each data log of F2FS to be written in parallel to multiple zones mapped to different chips to improve the inter-zone parallelism. First, a parallel zone grouping method is proposed to group zones mapped to different chips into multiple parallel zone groups. Then, a parallel zone group allocation method is given to allocate different parallel zone groups to data logs with different hotness. After that, a parallel zone group write mechanism is designed to allocate the space of multiple zones in a parallel zone group at the stripe size of a zone, allowing a request to be written to multiple zones in parallel. Extensive experiments based on an NVMe SSD emulator (<em>ConfZNS</em>) show <em>FPZone</em> can effectively improve the utilization of the chip, I/O latency and the performance of F2FS compared to the baseline scheme.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108044"},"PeriodicalIF":6.2,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144779296","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Online Learning from data streams via decentralized and asynchronous SGD","authors":"Mauro Dalle Lucca Tosi ,&nbsp;Martin Theobald","doi":"10.1016/j.future.2025.108052","DOIUrl":"10.1016/j.future.2025.108052","url":null,"abstract":"<div><div>Online Learning (OL) is a sub-field of Machine Learning (ML) which focuses on solving time-sensitive problems through iterative learning from data streams. This emerging field is characterized by the challenge of <em>concept drifts</em>, where the underlying distribution of the incoming data values evolves over time. Traditional OL algorithms, while efficient and less resource-intensive than conventional ML methods, often fall short in solving non-linear, high-dimensional problems. This prevalent gap has recently led to the integration of Artificial Neural Networks (ANN) into OL settings. These models support real-time inference. However, because they rely on offline training, their performance often degrades during or shortly after concept drifts. In this paper, we extend TensAIR, an online stream-processing engine that we specifically designed for the distributed training of ANN models. Our extensions allow TensAIR to automatically identify concept drifts using the OPTWIN drift detector algorithm, triggering the retraining of the ANN models as soon as drifts are detected. Additionally, we propose a novel <em>decentralized and asynchronous stochastic gradient descent</em> (DASGD) algorithm, which is central to TensAIR’s performance improvements over existing methods, and we formally prove its convergence under the specified conditions. We assessed TensAIR both in single-server and HPC settings, evaluating its distributed training performance over various multi-CPU and multi-GPU scenarios. As result, we show TensAIR to converge within the best known theoretical bounds while achieving up to <span><math><mrow><mn>78</mn><mo>×</mo></mrow></math></span> higher sustainable throughput than state-of-the-art baselines. Based on our results, we expect to inspire further research and applications exploiting the distributed training of ANN models in HPC platforms for a wide range of OL settings.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108052"},"PeriodicalIF":6.2,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144860856","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Distributed machine learning based on quantum cloud with quantum homomorphic encryption","authors":"Lin Zeng ,&nbsp;Yan Chang ,&nbsp;Xuejian Zhang ,&nbsp;Weifeng Xue ,&nbsp;Shibin Zhang ,&nbsp;Lili Yan ,&nbsp;Zhijian Gou","doi":"10.1016/j.future.2025.108053","DOIUrl":"10.1016/j.future.2025.108053","url":null,"abstract":"<div><div>In the era of Noisy Intermediate-Scale Quantum (NISQ) technology, variational quantum algorithms have emerged as a prominent method to showcase quantum superiority. Quantum federated learning (QFL), employing these algorithms in a distributed computing setting, enhances training performance and protects user data privacy. However, this approach significantly increases the demand for quantum capabilities, which are often beyond the reach of average users. This paper introduces a novel approach to QFL that incorporates quantum homomorphic encryption to enhance data security and privacy during collaborative training. We present the Quantum Circuit Random Reconstruction Homomorphic Encryption Algorithm (QCRRA), designed to facilitate secure and efficient model training over quantum networks without compromising data privacy. The QCRRA allows participants with no quantum capabilities to engage in federated learning by encrypting their quantum circuits, thus obviating direct exposure of sensitive data to quantum servers. We analyze the performance of our proposed approach through binary classification tasks on both MNIST, CIFAR10 and ad-hoc datasets, demonstrating that QCRRA maintains the integrity and accuracy of the learning process while significantly reducing the risk of data leakage. This study not only underscores the viability of QFL under stringent privacy constraints but also sets a precedent for future research in secure, decentralized quantum machine learning frameworks.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108053"},"PeriodicalIF":6.2,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144830717","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"OMP4Py: A pure Python implementation of openMP","authors":"César Piñeiro,&nbsp;Juan C. Pichel","doi":"10.1016/j.future.2025.108035","DOIUrl":"10.1016/j.future.2025.108035","url":null,"abstract":"<div><div>Python demonstrates lower performance in comparison to traditional high performance computing (HPC) languages such as C, C++, and Fortran. This performance gap is largely due to Python’s interpreted nature and the Global Interpreter Lock (GIL), which hampers multithreading efficiency. However, the latest version of Python includes the necessary changes to make the interpreter thread-safe, allowing Python code to run without the GIL. This important update will enable users to fully exploit multithreading parallelism in Python. In order to facilitate that task, this paper introduces OMP4Py, the first pure Python implementation of OpenMP. We demonstrate that it is possible to bring OpenMP’s familiar directive-based parallelization paradigm to Python, allowing developers to write parallel code with the same level of control and flexibility as in C, C++, or Fortran. The experimental evaluation shows that OMP4Py significantly impacts the performance of various types of applications, although the current threading limitations of Python’s interpreter (v3.13) reduce its effectiveness for numerical applications.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"175 ","pages":"Article 108035"},"PeriodicalIF":6.2,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144773127","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}