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

{"title":"TPQA: Efficient attention architecture with task-aware pattern-guided quantization","authors":"Sijia Wang ,&nbsp;Shengbing Zhang ,&nbsp;Lun Zhang ,&nbsp;Yichao Yuan ,&nbsp;Yawen Zhao ,&nbsp;Xinyu Zhang ,&nbsp;Meng Zhang","doi":"10.1016/j.future.2025.108352","DOIUrl":"10.1016/j.future.2025.108352","url":null,"abstract":"<div><div>Attention mechanisms have become a cornerstone of modern deep learning models, yet their computational intensity poses significant deployment challenges for resource-limited devices. While quantization offers a potential solution, current approaches typically employ uniform precision assignment schemes across all attention heads, neglecting critical variations in head-specific contributions across different tasks. This oversight results in substantial computational redundancy for those attention heads with fewer contributions, impacting overall performance. Through systematic analysis of head pattern characteristics in transformer models, we reveal two key insights: different attention heads exhibit distinct task-aware patterns, and their varying contributions to model performance directly dictate differentiated quantization demands across heads. Building on these findings, we propose TPQA, a novel algorithm and accelerator co-design architecture for efficient deployment of transformer models. TPQA strategically assigns adaptive precision levels to each head based on pre-identified patterns, thereby reducing computational overhead while preserving model accuracy. Furthermore, TPQA employs a data reordering strategy to transform irregular workloads into structured formats and introduces a dedicated accelerator with an attention-weights-stationary dataflow to efficiently process these structured workloads. Comprehensive evaluations demonstrate TPQA’s superior performance, achieving up to 2.1 ×  speedup and 3.4 ×  energy efficiency improvement over state-of-the-art accelerators while maintaining &lt;1% accuracy loss on various tasks.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108352"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145893776","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":"Topology-aware virtual machine placement for improving cloud servers resource utilization","authors":"Donglai Ma ,&nbsp;Xiaoyu Cao ,&nbsp;Jianchen Hu ,&nbsp;Tianyi Xia ,&nbsp;Yuzhou Zhou ,&nbsp;Kang Liu ,&nbsp;Lei Zhu ,&nbsp;Li Su ,&nbsp;Feng Gao","doi":"10.1016/j.future.2025.108361","DOIUrl":"10.1016/j.future.2025.108361","url":null,"abstract":"<div><div>As cloud computing offers increasingly sophisticated services, the optimal decisions of virtual machine (VM) placement become rather complicated, which may significantly influence the efficiency and profitability of cloud data centers (CDCs). In this paper, a realistic and holistic resource allocation model is proposed for the VM placement problem. Both the multi-layer topology of CDCs with complex topology-related user requests and the impact of multi-NUMA structures within servers are incorporated. Moreover, a novel objective function is developed to maximize overall resource utilization over an extended time horizon. The remaining resources of a server are characterized through the provision of three types of value: the value of hosting the current VM request, the potential value of accommodating future VM requests, and the topological value. A sophisticated value function is designed to integrate these components and quantify the overall benefit of placing VMs on a server, accounting for both the present and future values. As the resulting integer programming (IP) formulation is essentially NP-hard, a value-driven online algorithm is customized and developed to make online placement decisions following the proposed value function. By sequentially assigning VMs to feasible servers that maximize the evaluated placement value, our algorithm achieves a desirable trade-off between solution quality and computational efficiency. Numerical experiments on a practical cloud computing dataset demonstrate the effectiveness, efficiency, and scalability of the proposed VM placement method. Our test results indicate that the online placement decisions achieve over 80% of the global optimum (i.e., obtained from offline optimization), which outperforms other popular online methods, e.g., Fit-class heuristics and Deep Q-Network (DQN) based learning method. Besides, even with a challenging scale of 30,000 servers, our algorithm can make efficient placement decisions for 100 VMs within 1 s.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108361"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145893748","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":"On-device explainable artificial intelligence for the semantic web of everything","authors":"Davide Loconte ,&nbsp;Saverio Ieva ,&nbsp;Grazia Mascellaro ,&nbsp;Agnese Pinto ,&nbsp;Giuseppe Loseto ,&nbsp;Floriano Scioscia ,&nbsp;Michele Ruta","doi":"10.1016/j.future.2025.108310","DOIUrl":"10.1016/j.future.2025.108310","url":null,"abstract":"<div><div>As the Internet of Things (IoT) evolves into an Internet of Everything (IoE), adapting Artificial Intelligence (AI) and Machine Learning (ML) approaches to pervasive computing devices is not enough. Collaborative intelligence is required, calling for on-device AI frameworks combining adequate accuracy and computational efficiency levels with incremental learning on continuous data streams, federated learning in distributed architectures and symbolic explainability formalisms to foster trustworthiness with interpretable trained models and comprehensible prediction outcomes. To fill this gap, the paper introduces a five-star rating for on-device AI based on the Semantic Web of Everything (SWoE) paradigm, and presents the five-star <span>Mafalda</span> 2.0 framework. It combines statistical data processing with Knowledge Graph technologies for information representation and automated reasoning to support: semi-automatic or fully data-driven ontology definition; on-device training to generate highly interpretable semantics-based models; prediction framed as a semantic matchmaking problem, exploiting non-standard reasoning services endowed with logic-based justifications to provide comprehensible results as well as counterfactual and contrastive explanations. An experimental campaign on four publicly available datasets has been carried out to validate the efficiency and accuracy of the proposal, along with federated learning and explainability examples.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108310"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145893752","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":"A Portable Compiler-Runtime Approach for Scalability Prediction","authors":"Nicolai Stawinoga ,&nbsp;Sohan Lal ,&nbsp;Biagio Cosenza ,&nbsp;Philip Salzmann ,&nbsp;Peter Thoman ,&nbsp;Thomas Fahringer","doi":"10.1016/j.future.2025.108337","DOIUrl":"10.1016/j.future.2025.108337","url":null,"abstract":"<div><div>Highly scalable parallel applications can efficiently solve expensive computational problems when run on a large number of compute nodes. However, selecting the optimal number of nodes for a compute job of a given size is non-trivial, and allocating too few or too many nodes may not yield the expected performance. Knowing the scaling behavior of an application in advance enables us, for example, to make optimal use of the available hardware resources. We introduce a novel, portable approach to predict the scalability of parallel applications written in modern high-level programming models. We propose a predictive compiler-runtime framework based on Celerity, a task-based distributed runtime system that enables executing SYCL codes on clusters. The framework targets a broad range of computing systems, from CPU to GPU clusters, and proposes a model that combines machine learning, communication modeling and DAG heuristics. Experimental results on two large-scale clusters, JUWELS and Marconi-100, show accurate scalability prediction of unseen single and multi-task applications.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108337"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145845129","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":"Dispatching advanced and adaptive intrusion responses for IIoT-based systems","authors":"Jacobo Elicha,&nbsp;Javier Lopez","doi":"10.1016/j.future.2025.108314","DOIUrl":"10.1016/j.future.2025.108314","url":null,"abstract":"<div><div>The ever-increasing number of cyber-attacks poses a serious challenge to incident response teams. Recent cyber-attacks, such as the attack against energy distribution companies in Ukraine, highlight the disruption which can be caused and its consequences. More than 53 % of recorded incidents targeted essential entities, which heavily rely in Industrial Internet of Things (IIoT) devices, according to ENISA. Despite the amount of work in Cyber Threat Intelligence (CTI) and Intrusion Detection Systems (IDSs), automated response systems have been avoided in connected industrial environments mainly due to the criticality of the underlying assets, where a misstep has the potential to result in the disruption of critical processes. This paper therefore presents an Early and Adaptive Automated Intrusion Response Service for industrial environments, named EAIRS, which combines several techniques, including expert systems and reinforcement learning, to classify and mitigate anomalies detected by IDSs. The incidents EAIRS is designed to face range from network to host-based attacks. This paper provides the architecture for the described approach and the evaluation of a proof-of-concept implementation on an experimental testbed.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108314"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785033","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":"LLM+m: Dual-model chatGPT-based product training and testing with adversarial attack and defense","authors":"Ren-Hung Hwang ,&nbsp;Yu-Hung Hsiao ,&nbsp;Ying-Dar Lin ,&nbsp;Yuan-Cheng Lai","doi":"10.1016/j.future.2025.108328","DOIUrl":"10.1016/j.future.2025.108328","url":null,"abstract":"<div><div>Large language models (LLMs) like ChatGPT are increasingly leveraged in diverse applications. This study explores the integration of ChatGPT with machine learning models (LLM + m) to develop dual-model systems for intrusion detection–covering host-based (HIDS) and network-based (NIDS) systems–and image classification. By employing ChatGPT for data preprocessing and model generation, we evaluated the systems’ robustness against a range of adversarial attacks, including hypnotic attacks targeting the LLM, traditional adversarial attacks on ML models, and combined attacks affecting both. To counter these threats, we implemented adversarial training, ensemble models, and robustness prompts, significantly enhancing system resilience. Experimental results showed that combined attacks caused F1 score drops of up to 50 %, exposing critical vulnerabilities in dual-model systems. However, the proposed defenses reduced these losses to approximately 5 %, demonstrating their effectiveness in securing LLM-based dual-model systems against increasingly sophisticated adversarial threats.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108328"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785054","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":"Leveraging cutting-edge high performance computing for large-scale applications","authors":"Claude Tadonki ,&nbsp;Gabriele Mencagli ,&nbsp;Leonel Sousa","doi":"10.1016/j.future.2026.108374","DOIUrl":"10.1016/j.future.2026.108374","url":null,"abstract":"<div><div>High Performance Computing (HPC) recently entered into the exascale era, marking an important milestone of its history. High-end supercomputers and clusters with remarkable levels of performance are now commonly available for general and specific computational needs, thereby increasing the focus on HPC and related topics. Leveraging the potential of high-speed processing units is an HPC skillful task that requires in-depth knowledge in both hardware and software domains. In fact, the architectural structure of cutting-edge HPC processors is complex and involves several specialized features provided through specific units/mechanisms, the processing constraint/overhead of which can turn out to be an efficiency bottleneck. Large-scale supercomputers present greater challenges due to the significant overhead associated with interprocessor communication and synchronization. The evolution of HPC appears closely tied to the growing demand for speed from large-scale applications like complex combinatorial problems, big data applications, the training of large-scale AI models and high-precision simulations, to name a few. As a result, the implementation of cutting-edge techniques should remain scalable on large-scale machines for the benefit of end-users.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108374"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145962155","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":"SP-IFDA-Traj: Optimizing differentially private trajectory publishing for enhanced utility","authors":"Yongxin Zhao ,&nbsp;Hao Lin ,&nbsp;Xiangtian Zheng ,&nbsp;Yixuan Song ,&nbsp;Jinze Du","doi":"10.1016/j.future.2025.108359","DOIUrl":"10.1016/j.future.2025.108359","url":null,"abstract":"<div><div>Differential privacy (DP) is crucial for trajectory data publication, yet the utility of existing prefix-tree-based mechanisms heavily depends on intricate parameter tuning, often yielding suboptimal performance. To address this, we propose SP-IFDA-Traj, a novel framework for automatically and jointly optimizing critical parameters in personalized noisy prefix-tree trajectory publishing. SP-IFDA-Traj enhances the Flow Direction Algorithm (FDA) for optimizing parameters by: 1) employing chaotic mapping initialization for improved population diversity, 2) incorporating adaptive neighborhood generation to balance exploration and exploitation, and 3) leveraging Spark-based parallelized fitness evaluation for enhanced efficiency. Guided by a meticulously designed fitness function targeting maximal data utility, our framework optimizes Hilbert encoding order, privacy budget allocation, and pruning strategies. Extensive experiments on real-world large-scale trajectory datasets demonstrate that SP-IFDA-Traj substantially improves the privacy-utility trade-off. In convergence tests, the optimization engine improves the average fitness over FDA by 1.74% on BJ-Day3 and 103.98% on BJ-Day7, indicating consistently superior convergence across heterogeneous datasets. In terms of trajectory query accuracy, SP-IFDA-Traj reduces error to about 1% of that of baseline methods and approximately 10% of that of other existing optimization strategy models.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108359"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145897246","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":"COZO : A secure and efficient blockchain-enhanced federated learning paradigm with optimized storage and equitable contribution valuation","authors":"Ning Liu ,&nbsp;Yizhi Zhou ,&nbsp;Yuchen Qin ,&nbsp;Qinzheng Feng ,&nbsp;Heng Qi","doi":"10.1016/j.future.2025.108362","DOIUrl":"10.1016/j.future.2025.108362","url":null,"abstract":"<div><div>Federated Learning (FL) serves as a crucial enabler for digital supply chain transformation by supporting secure multi-party data collaboration while preserving privacy. However, its real-world implementation encounters two critical limitations: (1) conventional contribution assessment approaches (e.g., Shapley value) often ignore individual rationality, resulting in biased incentives that undermine long-term engagement; and (2) the underlying blockchain storage infrastructure suffers from inherent performance constraints, including slow read/write operations that cannot keep pace with FL’s need for real-time data exchange and frequent model updates. To overcome these challenges, we propose a framework that leverages Clique Games Solution theory to ensure fair and efficient contribution measurement. Our solution incorporates coalition screening and Pareto optimization to guarantee individual rationality with polynomial computational complexity. Additionally, we integrate COLE-a columnar storage module enhanced with learned indexes and LSM-tree merge mechanisms-to dramatically accelerate data access. Experimental results demonstrate that our approach achieves 5.72e-4 reduction in core distance for contribution assessment stability, improves storage IOPS by 28.7%, and maintains model accuracy within 1.2% of the baseline methods, outperforming mainstream baselines in evaluation stability and storage performance while maintaining competitive model accuracy, thus offering a comprehensive and practical solution for privacy-aware distributed learning in sensitive environments.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108362"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145902511","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":"A heuristic approach to Spark workflow task scheduling on heterogeneous nodes","authors":"Mehboob Hussain ,&nbsp;Ying Xu ,&nbsp;Zeeshan Abbas ,&nbsp;Ali Kamran ,&nbsp;Amir Rehman ,&nbsp;Muhammad Yasir","doi":"10.1016/j.future.2025.108335","DOIUrl":"10.1016/j.future.2025.108335","url":null,"abstract":"<div><div>Big data applications often use workflows shown as Directed Acyclic Graphs (DAGs). These DAGs link jobs and stages with order rules. Regular Spark-based workflow scheduling assumes all executors are identical and expects jobs to run one after another. This approach fails in mixed cloud setups, where machines vary in speed and computing capabilities. Efficient workflow scheduling in heterogeneous Spark nodes is crucial for reducing makespan and achieving load balancing. However, many existing methods overlook the challenge created by DAG-constrained jobs, stage-level dependencies, and varied node performance. This paper addresses efficient workflow scheduling on a mixed Spark cluster. We seek to minimize workflow completion time while keeping node loads balanced. We propose a modified Spark framework. It includes both a job scheduler and a stage scheduler tailored for mixed setups. Our method introduces multi-level node classification based on load status, a Speculative Stage Execution strategy for dynamic scheduling, and a Node Awareness Strategy for real-time task assignment. Compared to Rainbow, SAF, and DSWTS algorithms, SWTS reduces makespan by up to 40%, improves load balancing by 55%, and increases resource utilization by 20%. This demonstrates superior efficiency across all workflows.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"179 ","pages":"Article 108335"},"PeriodicalIF":6.2,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145823158","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}