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

{"title":"Causal invariant geographic network representations with feature and structural distribution shifts","authors":"Yuhan Wang ,&nbsp;Silu He ,&nbsp;Qinyao Luo ,&nbsp;Hongyuan Yuan ,&nbsp;Ling Zhao ,&nbsp;Jiawei Zhu ,&nbsp;Haifeng Li","doi":"10.1016/j.future.2025.107814","DOIUrl":"10.1016/j.future.2025.107814","url":null,"abstract":"<div><div>Relationships between geographic entities, including human-land and human-people relationships, can be naturally modelled by graph structures, and geographic network representation is an important theoretical issue. The existing methods learn geographic network representations through deep graph neural networks (GNNs) based on the i.i.d. assumption. However, the spatial heterogeneity and temporal dynamics of geographic data make the out-of-distribution (OOD) generalisation problem particularly salient. We classify geographic network representations into invariant representations that always stabilise the predicted labels under distribution shifts and background representations that vary with different distributions. The latter are particularly sensitive to distribution shifts (feature and structural shifts) between testing and training data and are the main causes of the out-of-distribution generalisation (OOD) problem. Spurious correlations are present between invariant and background representations due to selection biases/environmental effects, resulting in the model extremes being more likely to learn background representations. The existing approaches focus on background representation changes that are determined by shifts in the feature distributions of nodes in the training and test data while ignoring changes in the proportional distributions of heterogeneous and homogeneous neighbour nodes, which we refer to as structural distribution shifts. We propose a feature-structure mixed invariant representation learning (FSM-IRL) model that accounts for both feature distribution shifts and structural distribution shifts. To address structural distribution shifts, we introduce a sampling method based on causal attention, encouraging the model to identify nodes possessing strong causal relationships with labels or nodes that are more similar to the target node. This approach significantly enhances the invariance of the representations between the source and target domains while reducing the dependence on background representations that arise by chance or in specific patterns. Inspired by the Hilbert–Schmidt independence criterion, we implement a reweighting strategy to maximise the orthogonality of the node representations, thereby mitigating the spurious correlations among the node representations and suppressing the learning of background representations. In addition, we construct an educational-level geographic network dataset under out-of-distribution (OOD) conditions. Our experiments demonstrate that FSM-IRL exhibits strong learning capabilities on both geographic and social network datasets in OOD scenarios.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107814"},"PeriodicalIF":6.2,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143677904","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":"Chained continuous quantum federated learning framework","authors":"Dev Gurung,&nbsp;Shiva Raj Pokhrel","doi":"10.1016/j.future.2025.107800","DOIUrl":"10.1016/j.future.2025.107800","url":null,"abstract":"<div><div>The integration of quantum machine learning into federated learning paradigms is poised to transform the future of technologies that depend on diverse machine learning methodologies. This research delves into Quantum Federated Learning (QFL), presenting an initial framework modeled on the Federated Averaging (FedAvg) algorithm, implemented via Qiskit. Despite its potential, QFL encounters critical challenges, including (i) susceptibility to a single point of failure, (ii) communication bottlenecks, and (iii) uncertainty in model convergence. Subsequently, we dive deeper into QFL and propose an innovative alternative to traditional server-based QFL. Our approach introduces a chained continuous QFL framework (ccQFL), which eliminates the need for a central server and the FedAvg method. In our framework, clients engage in a chained continuous training process, where they exchange models and collaboratively enhance each other’s performance. This approach improves both the efficiency of communication and the accuracy of the training process. Our experimental evaluation includes a proof-of-concept to demonstrate initial feasibility and a prototype study simulating TCP/IP communication between clients. This simulation enables concurrent operations, verifying the potential of ccQFL for real-world applications. We examine various datasets, including Iris, MNIST, synthetic and Genomic, covering a range of data sizes from small to large. For further validity of our proposed method, we extend our experimental analysis in other frameworks such as PennyLane and TensorCircuit where we include various ablation studies covering major considerations and factors that impact the framework to study validity, robustness, practicality, and others. Our results show that the ccQFL framework achieves model convergence, and we evaluate other critical metrics such as performance and communication delay. In addition, we provide a theoretical analysis to establish and discuss many factors such as model convergence, communication costs, etc.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107800"},"PeriodicalIF":6.2,"publicationDate":"2025-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143677751","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An accurate and efficient self-distillation method with channel-based feature enhancement via feature calibration and attention fusion for Internet of Things","authors":"Qian Zheng ,&nbsp;Shengbo Chen ,&nbsp;Guanghui Wang ,&nbsp;Linfeng Li ,&nbsp;Shuo Peng ,&nbsp;Zhonghao Yao","doi":"10.1016/j.future.2025.107816","DOIUrl":"10.1016/j.future.2025.107816","url":null,"abstract":"<div><div>With the rise of the Internet of Things (IoT), using convolutional neural networks (CNNs) for image tasks on edge devices has become prevalent, but the increased size and complexity of neural networks for better performance is not ideal for resource-limited embedded devices. Self-distillation, which does not need a pre-trained complex model, has been introduced to utilize knowledge distillation during the model’s own training, thus enhancing performance. However, the model accuracy and efficiency of current self-distillation techniques still need investigation to meet real-world demands in IoT scenarios. Therefore, this paper proposes an improved self-distillation with Channel-Based Feature Enhancement (CBFE) via feature calibration and attention fusion, which improves network performance with minimal extra load. In particular, we first propose a channel-based feature calibration module. This module uses 1x1 convolutions to reduce and then restore the channel dimension of the neural network output feature maps. For each input feature map, it generates a new feature map, which is then element-wise multiplied with the original feature map to enhance representation. Second, we introduce a channel attention-based feature fusion network branch that refines a more accurate feature representation to better guide the training of shallow layers of the network. Experimental results show that our method surpasses the state-of-the-art methods, demonstrating enhanced performance and generalization on various benchmarks.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107816"},"PeriodicalIF":6.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143677752","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":"Improving database performance by leveraging network-assisted logging","authors":"Hwajung Kim","doi":"10.1016/j.future.2025.107785","DOIUrl":"10.1016/j.future.2025.107785","url":null,"abstract":"<div><div>In mission-critical systems like databases and transaction processing systems, write-ahead logging (WAL) is commonly used to ensure fault tolerance against power failures and system malfunctions. However, WAL requires data to be logged before it is permanently stored in persistent storage, causing delays that can slow down the system, even when using advanced technologies like Optane persistent memory. Although seemingly small, such delays can accumulate and affect overall transaction performance. In this paper, we propose an in-transit logging (<em>ITLogging</em>) scheme that performs logging at the network layer by capturing important data upon its arrival at the destination system. Our scheme filters incoming packets and logs the necessary data from the payload before any processing occurs. In case of data loss, our scheme replays packet deliveries to the target system by mimicking the original client actions for recovery. We implement the proposed scheme by allocating a separate core for packet inspection, ensuring that logging operations are handled independently of the application layer’s data processing, thereby avoiding delays in the main processing flow. The experimental results demonstrate that our scheme improves database throughput by 16% for TPC-C and 15% for LinkBench on MySQL, compared with vanilla MySQL.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107785"},"PeriodicalIF":6.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143677753","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":"Delay-cost computation offloading for on-board emergency tasks in LEO Satellite Edge Computing networks","authors":"Changhao Li ,&nbsp;Zhenmou Liu ,&nbsp;Zhicong Ye ,&nbsp;Guoguang Wen ,&nbsp;Zong-Fu Luo ,&nbsp;Chuanfu Zhang","doi":"10.1016/j.future.2025.107797","DOIUrl":"10.1016/j.future.2025.107797","url":null,"abstract":"<div><div>The increasing computational capabilities of Low Earth Orbit (LEO) constellations have significantly augmented their autonomy and operational flexibility. Complex onboard tasks such as observation, sensing, and situational awareness can be processed and executed directly on the Satellite Edge Computing (SEC) networks. Due to the time-varying characteristics of inter-satellite links and the uncertainty in the load of edge satellites, efficient offloading of on-board tasks presents significant challenges. We introduce an on-board distributed task offloading method for LEO satellite tasks in emergency to enhance service quality. We initially design a dynamic offloading scheme, in which data-source satellites can transmit tasks to edge nodes. Then, we formulate the multi-hop satellite network dynamic offloading (MSNDO) problem to minimize system delay and maximize success ratio of time-sensitive tasks under multiple constraints. Finally, we propose a distributed deep reinforcement learning algorithm that allows individual satellites to design offloading strategies without knowing the decision-making patterns of other satellites. Simulation experiments show that the proposed algorithm can utilize the edge satellite processing capabilities more efficiently and significantly improve the performance of the SEC system.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107797"},"PeriodicalIF":6.2,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143677754","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":"DrugPred: An ensemble learning model based on ESM2 for predicting potential druggable proteins","authors":"Hong-Qi Zhang ,&nbsp;Shang-Hua Liu ,&nbsp;Jun-Wen Yu ,&nbsp;Rui Li ,&nbsp;Dong-Xin Ye ,&nbsp;Yan-Ting Jin ,&nbsp;Cheng-Bing Huang ,&nbsp;Ke-Jun Deng","doi":"10.1016/j.future.2025.107801","DOIUrl":"10.1016/j.future.2025.107801","url":null,"abstract":"<div><div>The Human Genome Project has generated abundant data for a long time, but transforming this data into practical and usable drug or drug target products remains challenging. This study proposed an ensemble learning model, called DrugPred, to achieve prediction of drug targets using evolutionary scale modeling (ESM2) and amino acid composition (AAC) as features. ESM2 utilized deep learning technology to study the sequence-structure-function relationship of protein sequences, extracting highly abstract features of proteins. AAC translated protein sequences into amino acid percentages, reflecting the composition of amino acids in proteins. The integration of two features constituted a multidimensional and diverse feature space, enabling the model to perform well in predicting drug targets. We input the fused features into four machine learning algorithms for separate training and generated the prediction probabilities, then input them into a support vector machine for voting decisions. This ensemble learning overcame the bias of a single algorithm model in information learning and improved the stability and accuracy of the model. After comprehensive evaluation, the model achieved an accuracy of 0.9691 with an area under the receiver operating characteristic curve (AUC) value of 0.9868. We also used t-distributed Stochastic Neighbor Embedding (t-SNE) and SHapley Additive exPlanations (SHAP) techniques to explore the interpretability of the DrugPred model. This study provided a fresh perspective and method for identifying drug targets, offering robust support for future drug development. We have developed and made publicly accessible a web server based on the DrugPred model. The web server is accessible at <span><span>http://drugpred.lin-group.cn/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"170 ","pages":"Article 107801"},"PeriodicalIF":6.2,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143704949","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":"Efficient memory management with co-timeout policy eviction and history-enlightened selective strategy installation","authors":"Xianfeng Li ,&nbsp;Haisheng Yu ,&nbsp;Xue Yang ,&nbsp;Haoran Sun ,&nbsp;Yan Huang","doi":"10.1016/j.future.2025.107799","DOIUrl":"10.1016/j.future.2025.107799","url":null,"abstract":"<div><div>In Software-Defined Network (SDN), the controller plays a crucial role in implementing fine-grained network policies by installing flow rules in the switch’s flow table. However, the limited capacity of the flow table, often implemented using TCAM, poses scalability challenges due to its low density and high energy consumption. To address this issue, this paper focuses on two key aspects: (1) early eviction of installed flow rules and (2) selective installation of flow rules.</div><div>To tackle the first aspect, we propose an adaptive timeout mechanism called Two-Stage Timeout (TST) that enhances the flow table architecture. TST enables the efficient eviction of short-lived flows, creating space for more valuable flow rules. In addition, the introduction of the Inactive Flow Queue (IFQ) improves the retention of active flows, enhancing overall table management.</div><div>For the second aspect, we introduce RICHRIN, a mechanism that combines historical and real-time information to prevent the installation of unnecessary flow rules that contribute little to cache hit rates. RICHRIN effectively filters out a significant portion of these unproductive flows, reducing pollution in the flow table.</div><div>To evaluate the performance of TST and RICHRIN, we conduct experiments using real network packet traces from CAIDA. The results demonstrate that these mechanisms significantly improve the rule cache hit ratio and substantially reduce the number of rule installations. This research effectively addresses the scalability problem of the flow table and provides valuable insights for optimizing flow table management in SDN environments.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107799"},"PeriodicalIF":6.2,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143681392","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":"Effect of implementations of the N-body problem on the performance and portability across GPU vendors","authors":"Rodrigo A.C. Bartolomeu ,&nbsp;René Halver ,&nbsp;Jan H. Meinke ,&nbsp;Godehard Sutmann","doi":"10.1016/j.future.2025.107802","DOIUrl":"10.1016/j.future.2025.107802","url":null,"abstract":"<div><div>Since Aurora entered the <span><span>TOP500</span><svg><path></path></svg></span> list in November 2023, the top ten systems saw some shifts in the ratio of GPU vendors represented. With each vendor supplying their own preferred programming models for their hardware, it becomes relevant to compare the portability of these models on other hardware platforms. For the present paper we implemented the N-body problem with different optimizations using native and portable programming frameworks. For each of those we determined the best performing optimized version on one target architecture and compared the performance achieved for each platform.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107802"},"PeriodicalIF":6.2,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143681467","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Keyed watermarks: A fine-grained watermark generation for Apache Flink","authors":"Tawfik Yasser ,&nbsp;Tamer Arafa ,&nbsp;Mohamed ElHelw ,&nbsp;Ahmed Awad","doi":"10.1016/j.future.2025.107796","DOIUrl":"10.1016/j.future.2025.107796","url":null,"abstract":"<div><div>Big Data Stream processing engines, exemplified by tools like Apache Flink, employ windowing techniques to manage unbounded streams of events. Aggregating relevant data within Windows is important for event-time windowing due to its impact on result accuracy. A pivotal role in this process is attributed to watermarks, unique timestamps signifying event progression in time. Nonetheless, the existing watermark generation method within Apache Flink, operating at the input stream level, exhibits a bias towards faster sub-streams, causing the omission of events from slower counterparts. Our analysis determined that Apache Flink’s standard watermark generation approach results in an approximate 33% data loss when 50% of median-proximate keys experience delays. Furthermore, this loss exceeds 37% in cases where 50% of randomly selected keys encounter delays. In this paper, we introduce a pioneering approach termed <em>keyed watermarks</em> to address data loss concerns and enhance data processing precision to a minimum of 99% in most scenarios. Our strategy facilitates distinct progress monitoring by creating individualized watermarks for each sub-stream (key). Within our investigation, we delineate the essential architectural and API modifications requisite for integrating keyed watermarks while also highlighting our experience in navigating the expansion of Apache Flink’s extensive codebase. Moreover, we conduct a comparative evaluation between the efficacy of our approach and the conventional watermark generation technique concerning the accuracy of event-time tracking, the latency of watermark processing, and the growth of Flink’s maintained state.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107796"},"PeriodicalIF":6.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143620560","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 self-organized MoE framework for distributed federated learning","authors":"Jungjae Lee,&nbsp;Wooseong Kim","doi":"10.1016/j.future.2025.107798","DOIUrl":"10.1016/j.future.2025.107798","url":null,"abstract":"<div><div>Federated Learning (FL) has solved the problem of data silos by enabling multiple participants to cooperatively train a global model while ensuring data privacy; however, it is still a challenge to establish a Distributed Federated Learning (DFL) framework that naturally suffers from the heterogeneity of devices and datasets. Rather than conventional FL algorithms that combine client models for a single global model, a Mixture of Experts (MoE) based FL is an effective alternative that can admit individual features on each client dataset by partitioning the entire latent space. In this study, we introduce the Self-Organized MoE Framework (SOMFed), which enhances the DFL lifecycle under asynchronous updates and statistical challenges of datasets. Considering that nodes are assumed to lack label information in contrast to most of previous studies, aside from their class data, we propose the Model Assessment and Selection (MASS) algorithm for the SOMFed framework, leveraging self-supervised learning. It evaluates and chooses suitable experts for own unlabeled dataset by differentiating the performance of the representation layers among experts using Bayesian optimization and Conditional Loss Adjustment (CLA). The SOMFed exhibits superior performance in extensive experiments with different non-IID distributions and stragglers compared to FedAVG, FedAsync, SCAFFOLD, FedAT, and Adaptive Expert Models (AEM). In particular, it demonstrates robustness against pathological non-IID distribution on CIFAR10, achieving accuracy of 79.42%.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"169 ","pages":"Article 107798"},"PeriodicalIF":6.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143620557","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}