IEEE Transactions on Services Computing最新文献

{"title":"BIPPO: Budget-Aware Independent PPO for Energy-Efficient Federated Learning Services","authors":"Anna Lackinger;Andrea Morichetta;Pantelis A. Frangoudis;Schahram Dustdar","doi":"10.1109/TSC.2026.3713497","DOIUrl":"10.1109/TSC.2026.3713497","url":null,"abstract":"Federated Learning (FL) is a promising approach for training models in large-scale IoT systems, natively ensuring load distribution and privacy. However, FL is agnostic to infrastructure efficiency, which is a critical concern in resource-constrained environments. Although Reinforcement Learning (RL)-based solutions improve client selection in FL, they still overlook infrastructure challenges, such as resource constraints and client churn. Moreover, they often lack practical applicability, as they rarely guarantee generalizability and energy efficiency. To fill this gap, we propose BIPPO (Budget-aware Independent Proximal Policy Optimization), an energy-efficient multi-agent RL (MARL) solution that enhances FL performance. We empirically evaluate BIPPO on two image classification tasks in highly budget-constrained, non-IID settings, a challenging scenario for vanilla FL. BIPPO’s redesigned <inline-formula><tex-math>$epsilon$</tex-math></inline-formula>-greedy sampler increases accuracy, convergence speed, and the model’s consistency compared to non-RL mechanisms, traditional PPO, and IPPO. In addition, BIPPO consumes only a negligible share of the budget, which remains constant as the number of clients increases, resulting in a training overhead up to three orders of magnitude lower than that of single-agent RL (SARL). Overall, BIPPO delivers a performant, stable, scalable, and sustainable solution for client selection in IoT-FL.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3361-3372"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11610669","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148460825","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":"PDTE: Outsourced Private Decision Tree Evaluation for Large Trees With Sublinear Communication","authors":"Dingcheng Li;Kai He;Jia-Nan Liu;Anjia Yang;Xueqiao Liu;Jian Weng","doi":"10.1109/TSC.2026.3689639","DOIUrl":"https://doi.org/10.1109/TSC.2026.3689639","url":null,"abstract":"Decision trees are critical in domains like medical diagnosis and network security, but their evaluation often compromises privacy. Existing privacy-preserving methods either reduce accuracy or incur high overheads. To address these limitations, we propose an efficient outsourced Private Decision Tree Evaluation (PDTE) protocol under a two-cloud architecture, encoding the tree into a two-dimensional array and using Oblivious Selection (OS) for secure evaluation. Our PRF-based OS achieves sublinear communication, enabling large-scale tree evaluation. We also support private random forest evaluation and categorical feature handling. Experiments show up to 167× faster computation in LAN, more than 33× faster computation in MAN and 173× better communication efficiency.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3170-3183"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133687","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":"Guest Editorial: Special Issue on Privacy-Preserving Machine Learning Services","authors":"Xingliang Yuan;Kwok Yan Lam;Ronald Cramer;Xiaoning Liu","doi":"10.1109/TSC.2026.3733384","DOIUrl":"https://doi.org/10.1109/TSC.2026.3733384","url":null,"abstract":"","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3140-3143"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11731481","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133531","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":"Blockchain-Enabled Elastic Task Offloading and Migration Scheme in Edge Computing System","authors":"Qiang He;Tianyi Qiu;Xingwei Wang;Ammar Hawbani;Lianbo Ma;Keping Yu;Liang Zhao","doi":"10.1109/TSC.2025.3634215","DOIUrl":"10.1109/TSC.2025.3634215","url":null,"abstract":"Mobile edge computing (MEC), which improves the quality of experience of mobile devices by offloading tasks to edge servers, has become a promising paradigm. This method can effectively reduce computational latency and lower energy consumption. However, traditional task offloading and migration schemes only consider a single static scenario, making it difficult to provide users with elastic computing services in changing environments. Moreover, the quality and cost of the same task vary depending on the service level. To tackle this problem, we propose a reinforcement learning approach called LSTM-SAC, which integrates long short-term memory (LSTM) with soft actor-critic (SAC). LSTM-SAC provides elastic computing services to users in MEC systems by adaptively selecting models based on the current environment, thereby enhancing the flexibility of MEC systems. In addition, we deploy the blockchain on the MEC system to enhance its security. All users must authenticate their identities by running a smart contract deployed on the blockchain before offloading tasks. The experimental results show that our proposed method outperforms benchmark methods in terms of convergence and security.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3373-3386"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145759510","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":"AIAER-Mask: Safe Neuro-Symbolic Intent-Correct Routing for Event-Driven Microservices","authors":"Hossam Hassan;Shadia Sarhan;Ahmed Shalaby","doi":"10.1109/TSC.2026.3708565","DOIUrl":"10.1109/TSC.2026.3708565","url":null,"abstract":"Event-driven microservice systems route heterogeneous operations through rapidly changing service replicas. Conventional performance-oriented routing systems, including static routing, EWMA-based selection, Power-of-Two-Choices, and unconstrained reinforcement-learning policies, can select low-latency replicas that are semantically invalid for the requested operation. Such routing mistakes may preserve raw throughput while violating application intent, consistency, privacy boundaries, or update semantics. This paper presents AIAER-Mask, a safe neuro-symbolic routing framework for intent-correct event routing. AIAER-Mask separates safety from adaptation: a lightweight symbolic intent guard derives a hard action mask from a capability map, while an offline discrete Soft Actor-Critic policy selects among only admissible replicas using telemetry features. The symbolic mask is not a neural model; it is a deterministic safety layer that prevents invalid actions by construction under the active validated map. The evaluation covers multi-replica routing, ablations, masked versus unmasked baselines, failure injection, dynamic/adversarial scenarios, broker-delay and local region-latency emulation, hyperparameter sensitivity, training diagnostics, reproducibility reporting, and DeathStarBench SocialNetwork/HotelReservation cross-workload evaluation. Across both workloads, hard symbolic masking preserves 100% intent correctness and 0% invalid actions for masked variants under validated and actively enforced capability maps, while performance-only no-mask baselines violate semantic routing constraints heavily. The mask itself is low-overhead for heuristic routers, whereas most additional overhead in full AIAER-Mask comes from neural inference, telemetry/state processing, and feature handling. Strong masked heuristics such as Mask+EWMA and Intent-safe P2C remain highly competitive and sometimes outperform the learned policy in latency and regret. We therefore position AIAER-Mask as a safety-constrained neuro-symbolic framework, not as the fastest router in every regime. The current evidence supports instantiation across two single-region DeathStarBench workloads and local region-like latency profiles; broader production-trace, real multi-broker, and real multi-region validation remains future work.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3336-3346"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148401113","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":"FedAtlas: A Privacy-Enhanced Federated Multi-Task Learning Framework Driven by Semantic Communication for Dual Heterogeneity","authors":"Xinyu Fu;Jiaqi Zhu;Xiaoding Wang;Xu Yang;Xuechao Yang;Jiwu Shu","doi":"10.1109/TSC.2026.3685997","DOIUrl":"10.1109/TSC.2026.3685997","url":null,"abstract":"Federated Multi-Task Learning (FMTL) faces fundamental challenges arising from dual heterogeneity, including intrinsic task heterogeneity and communication challenges induced by heterogeneous system environments. Task heterogeneity often leads to negative knowledge transfer due to complex non-linear task relationships, while communication inefficiency constrains system scalability and training efficiency. Existing FMTL methods typically rely on shallow task relation modeling and overlook the structural nature of task dependencies. To address these challenges, we propose FedAtlas, a structure-aware and privacy-enhanced FMTL framework that explicitly constructs a global non-linear task relationship atlas through an adaptive structure inference module. Guided by the inferred atlas, FedAtlas adopts a structure-aware dual-level aggregation strategy to jointly optimize shared encoders and personalized decoders, effectively mitigating negative transfer while preserving task-specific specialization. Furthermore, FedAtlas integrates a semantic communication mechanism to alleviate communication inefficiency under heterogeneous system conditions. By leveraging client-specific vector quantization codebook, the proposed design enables compact semantic feature transmission while providing intrinsic privacy protection without explicit noise injection. Extensive experiments demonstrate that FedAtlas outperforms state-of-the-art FMTL methods, showing 4.82% improvement in performance and efficiency trade-off.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3255-3268"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147725602","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":"AdaFUN: Enhancing Federated Unlearning With Adaptive Local Step in Edge Computing","authors":"Zhenguo Ma;Yanjing Sun;Hongli Xu;Jianchun Liu;Bowen Wang;Kaiwen Dong;Zixuan He","doi":"10.1109/TSC.2026.3685363","DOIUrl":"10.1109/TSC.2026.3685363","url":null,"abstract":"Federated Unlearning (FU) has emerged as a pivotal paradigm for ensuring the “Right To Be Forgotten” in contemporary intelligent services of Edge Computing (EC). Nevertheless, existing FU schemes typically assign a uniform local step to all edge devices, which inevitably incurs non-negligible waiting time. Moreover, regarding the heterogeneous computational capacity and non-IID local data across devices, a uniform local step will exacerbate the degradation in both training performance and unlearning efficacy. To this end, we propose an adaptive FU framework, named AdaFUN, which dynamically adjusts local steps for heterogeneous devices. Since the configuration of local steps severely influences the convergence rate and is closely correlated to the residual memory of the target data, we first analyze the relationship between local step and training performance alongside unlearning efficacy from a theoretical perspective. On these bases, we formulate an optimization problem with the objective to minimize the convergence upper bound and residual memory of the target data. We design effective algorithms to solve this problem, which adaptively adjust the local steps of edge devices according to their computational capacity and data distribution, thereby enhancing both training performance and unlearning efficacy. We conduct extensive experiments on real-world datasets, and the experimental results demonstrate the effectiveness of the proposed framework.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3241-3254"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147725601","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":"Semi-Automated Threat Modeling of Cloud-Based Systems Through Extracting Software Architecture From Configuration and Network Flow","authors":"Nicholas Pecka;Lotfi Ben Othmane;Bharat Bhargava;Renée Bryce","doi":"10.1109/TSC.2026.3724971","DOIUrl":"10.1109/TSC.2026.3724971","url":null,"abstract":"Traditional threat modeling occurs during design, but cloud deployments introduce unanticipated threats—especially multi-stage attacks chaining vulnerabilities across trust boundaries. Existing security tools analyze components in isolation, cannot detect architectural threats from system composition, and cannot validate runtime behavior against configured policies. This gap leaves organizations vulnerable to attacks exploiting architectural weaknesses. This paper addresses this gap through a key innovation: automatically inferring system architecture from runtime observations to enable continuous threat modeling. Our methodology combines static configuration analysis with observed network flows to construct architecture graphs reflecting actual operational behavior, then applies systematic threat detection using platform-agnostic abstractions (components, domains, interfaces, access policies, flows). This enables consistent threat identification across bare metal, Kubernetes, and cloud infrastructure without manual diagram maintenance. We validate the methodology using two case studies: a supply-chain system (8 components) and a container registry (20 components), injecting 17 infrastructure and ML threat types at both scales. Results show detection of all 17 threat types (100%) across both case studies and platforms, yielding 49 threat instances on bare-metal (e.g., 13 exposed secrets, 8 missing resource limits), 47 on Kubernetes, and 71 on AWS for the supply-chain system; the 20-component registry produced 47 AWS threat instances, while existing security tools detected only 6-47% with zero ML threat coverage, confirming the necessity of runtime-aware, architecture-level threat analysis. The methodology maintains detection accuracy as system complexity increases (2.5× component growth, 100% detection sustained).","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3621-3633"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148755526","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":"Fine-Grained Power Allocation for Power-Tiered Pricing Based Wireless Charging Services","authors":"Lijie Xu;Xintong Jin;Yuanbiao Shang;Shunping Yang;Haipeng Dai;Jia Xu","doi":"10.1109/TSC.2026.3717220","DOIUrl":"10.1109/TSC.2026.3717220","url":null,"abstract":"In wireless rechargeable sensor networks, simultaneous data monitoring and RF wireless charging can degrade sensing accuracy, thereby necessitating constraints on charging duration. Furthermore, prevailing coarse-grained charging models miss the benefits of fine-grained, adjustable power for optimizing cost efficiency. This paper introduces a novel paid charging service model where Charging Service Provider (CSP) deploys charging infrastructure and users request charging services from CSP. Within this model, we propose a power-tiered pricing based fine-grained charging service mechanism where different charging power levels have corresponding unit prices. We study the NP-hard problem of finding a fine-grained charging schedule that meets all sensor charging demands within a fixed service duration constraint while minimizing the total service fee, termed the Fine-grained Economical Charging Scheduling based on Temporal-constraint (FECST) problem. By transforming FECST into an equivalent Charging Service Fee Reduction-amount Maximization (CSFRM) problem and leveraging atomic scheduling strategies, we prove CSFRM is equivalent to a Multi-dimensional Knapsack constrained Submodular Function Maximization (MKSFM) problem. For MKSFM, a polynomial-time multiplicative updates based algorithm with a provable approximation ratio is devised. Building on this, we further propose the Approximate Fine-grained Charging Scheduling (AFCS) algorithm to solve FECST, achieving a guaranteed approximation ratio. Simulations show our solution consistently outperforms baselines, significantly reducing costs.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3461-3478"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148595593","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":"Near-Optimal Differentially Private k-Center for Service Computing With Access Control","authors":"Chaoqi Jia;Longkun Guo;Zhigang Lu;Chao Chen;Kok-Leong Ong","doi":"10.1109/TSC.2026.3686829","DOIUrl":"10.1109/TSC.2026.3686829","url":null,"abstract":"Many applications in service computing systems, such as location-based services (LBS), use the <inline-formula><tex-math>$k$</tex-math></inline-formula>-center problem to guide service placement and improve the quality of service. In such settings, access control (AC) is commonly employed to protect client privacy. However, as we demonstrate in this article, privacy leakage can still occur even in the presence of AC. To address this issue, we study the AC-<inline-formula><tex-math>$k$</tex-math></inline-formula>-center problem under differential privacy (DP), which has emerged as a leading paradigm for mitigating privacy risks by injecting calibrated noise while providing guarantees that remain robust against adversaries with arbitrary prior knowledge. We first propose a framework that integrates access control into the service placement process, thereby establishing a balance between privacy preservation and solution utility. Building on this framework, we present a novel algorithm for the differentially private AC-<inline-formula><tex-math>$k$</tex-math></inline-formula>-center problem that achieves the best possible 2-approximation ratio with an additive error. By leveraging intermediate structures from the continuous <inline-formula><tex-math>$k$</tex-math></inline-formula>-center problem and Voronoi diagrams, our approach reduces the additive error in the approximation ratio compared with state-of-the-art methods. Finally, we conduct extensive experiments on three real-world geographical datasets, demonstrating that our algorithm outperforms existing baselines in both solution quality and runtime efficiency.","PeriodicalId":13255,"journal":{"name":"IEEE Transactions on Services Computing","volume":"19 5","pages":"3322-3335"},"PeriodicalIF":6.2,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147735601","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}