Journal of Systems Architecture最新文献

{"title":"HPA: Manipulating deep reinforcement learning via adversarial interaction","authors":"Kanghua Mo ,&nbsp;Zhengxin Zhang ,&nbsp;Yuanzhi Zhang ,&nbsp;Yucheng Long ,&nbsp;Zhengdao Li","doi":"10.1016/j.sysarc.2026.103685","DOIUrl":"10.1016/j.sysarc.2026.103685","url":null,"abstract":"<div><div>Recent studies have demonstrated that policy manipulation attacks on deep reinforcement learning (DRL) systems can lead to the learning of abnormal policies by victim agents. However, existing work typically assumes that the attacker can manipulate multiple components of the training process, such as reward functions, environment dynamics, or state information. In IoT-enabled smart societies, where AI-driven systems operate in interconnected and data-sensitive environments, such assumptions raise serious concerns regarding security and privacy. This paper investigates a novel policy manipulation attack in competitive multi-agent reinforcement learning under significantly weaker assumptions, where the attacker only requires access to the victim’s training settings and, in some cases, the learned policy outputs during training. We propose the honeypot policy attack (HPA), in which an adversarial agent induces the victim to learn an attacker-specified target policy by deliberately taking suboptimal actions. To this end, we introduce a honeypot reward estimation mechanism that quantifies the amount of reward sacrifice required by the adversarial agent to influence the victim’s learning process, and adapts this sacrifice according to the degree of policy manipulation. Extensive experiments on three representative competitive games demonstrate that HPA is both effective and stealthy, exposing previously unexplored vulnerabilities in DRL-based systems deployed in IoT-driven smart environments. To the best of our knowledge, this work presents the first policy manipulation attack that does not rely on explicit tampering with internal components of DRL systems, but instead operates solely through admissible adversarial interactions, offering new insights into security challenges faced by emerging AIoT ecosystems.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103685"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146090348","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":"When fixes teach: Repair-aware contrastive learning for optimization-resilient binary vulnerability detection","authors":"Zhenzhou Tian ,&nbsp;Jiale Zhao ,&nbsp;Ming Fan ,&nbsp;Jiaze Sun ,&nbsp;Yanping Chen ,&nbsp;Lingwei Chen","doi":"10.1016/j.sysarc.2026.103722","DOIUrl":"10.1016/j.sysarc.2026.103722","url":null,"abstract":"<div><div>Deep learning (DL)-based vulnerability detection in source code are prevalent, yet detecting vulnerabilities in binary code using this paradigm remains underexplored. The few works typically treat input instructions as individual entities, failing to extract and leverage fine-grained information due to their inability to account for the inherent connections and correlations between code segments and the impact of compilation optimizations. To address these challenges, this paper proposes <strong><span>Delta</span></strong>, a novel approach that incorporates <strong>D</strong>ynamic contrastive l<strong>E</strong>arning with vu<strong>L</strong>nerabili<strong>T</strong>y repair <strong>A</strong>wareness to fine-tune pre-trained models, significantly enhancing the accuracy and efficiency of vulnerability detection in binary code. <span>Delta</span> proceeds by standardizing assembly instructions and utilizing function pairs that represent code before and after vulnerability repair along with their versions compiled under different optimization settings as contrastive learning samples. Building on these rich and diverse training signals, <span>Delta</span> fine-tunes CodeBERT using contrastive learning augmented with masked language modeling, resulting in a feature encoder CMBERT, which is adept at capturing nuanced vulnerability patterns in binary code and remain resilient to the impacts of compilation optimizations. DELTA is evaluated on the Juliet Test Suite dataset, achieving an average performance improvement of 8.04% in detection accuracy and 7.13% in F1 score compared to alternative methods.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103722"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146090397","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":"Comments on “Contention-aware workflow scheduling on heterogeneous computing systems with shared buses”","authors":"Rajesh Devaraj","doi":"10.1016/j.sysarc.2026.103700","DOIUrl":"10.1016/j.sysarc.2026.103700","url":null,"abstract":"<div><div>Heterogeneous computing systems (HCSs) use different types of processors to balance performance and efficiency for complex applications like workflows. These processors often share a communication bus. When multiple parts of an application try to send data at the same time, this shared bus gets congested, causing delays. Despite this being a common problem, few studies have looked at how to handle this communication bottleneck. To solve this, a new method called <em>Contention-Aware Clustering-based List scheduling</em> (CACL) is proposed. The objective of <em>CACL</em> is to minimize the overall schedule length for the input workflow application modeled as a Directed Acyclic Graph (DAG) to be executed on a HCS interconnected via shared communication buses. While solving this problem, <em>CACL</em> first assigns priorities to task nodes. However, this task priority assignment may occasionally lead to situations where a task is erroneously assigned higher priority compared to one or more of its predecessor tasks in the task graph. Since tasks are selected for processor assignment in the order of their priorities, this situation subsequently leads to violation of precedence relationships between tasks. In this comment, we present a counter example to highlight the design flaw in the task prioritization scheme and discuss possible ways to fix this flaw.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103700"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146090398","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":"Unmixing gradients: Uncovering persistent leakage in Federated Learning via Independent Component Analysis","authors":"Minhao Li ,&nbsp;Le Wang ,&nbsp;Zhaohua Li ,&nbsp;Rongxin Hu ,&nbsp;Tang Zhou ,&nbsp;Binxing Fang","doi":"10.1016/j.sysarc.2025.103681","DOIUrl":"10.1016/j.sysarc.2025.103681","url":null,"abstract":"<div><div>Federated Learning (FL), a cornerstone of privacy-preserving technology in the Artificial Intelligence of Things (AIoT) era, enables collaborative model training across edge devices via gradient sharing. However, Gradient Leakage Attacks (GLA) can reconstruct private data from these gradients, fundamentally undermining FL’s privacy guarantees. While recent Generation-based GLAs demonstrate significant promise in efficiency and quality, their core feature separation algorithms rely heavily on idealized properties of early-stage training. This dependency causes attack performance to decay rapidly as training progresses and fails in challenging scenarios, such as batches with duplicate labels. To overcome this, we reframe feature separation as a Blind Source Separation (BSS) problem, solved using Independent Component Analysis (ICA). Through a systematic analysis of the entire training lifecycle, we uncover the adversarial dynamics of key mathematical properties that govern BSS solvability. Based on these insights, we introduce a novel framework: <strong>ICA-driven Generative Attacks (ICA-GA)</strong>. Extensive experiments show that ICA-GA significantly outperforms baselines throughout the training lifecycle and exhibits remarkable robustness against challenging conditions, including batches with full label duplication, FedAVG, and gradient compression. Furthermore, our incremental generator fine-tuning strategy reduces the marginal cost of continuous multi-round attacks by an order of magnitude, making such threats highly practical. Our work reveals that the privacy risk in FL is far more persistent and severe than previously understood. Our code is publicly available at <span><span>https://github.com/liminhao-99/ICA-GA</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103681"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145941384","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":"Enhancing deterministic transmission in Time-Sensitive Networking: A Joint Guard Band Compression and Non-Disruptive Frame Preemption model","authors":"Ling Zheng ,&nbsp;Jingzhuo Liu ,&nbsp;Keyao Zhang ,&nbsp;Qianxi Men ,&nbsp;Li Zhen ,&nbsp;Weitao Pan","doi":"10.1016/j.sysarc.2026.103697","DOIUrl":"10.1016/j.sysarc.2026.103697","url":null,"abstract":"<div><div>In the Industrial Internet of Things (IIoT), the real-time interaction of critical control and sensing data imposes stringent requirements on network determinism and reliability. Time-Sensitive Networking (TSN) is widely utilized to ensure low latency and high-precision synchronization; however, efficiently coordinating the nanosecond-level determinism of Time-Triggered (TT) traffic with the high throughput of Event-Triggered (ET) traffic remains a core challenge. Existing Guard Band mechanisms lead to significant bandwidth waste due to fixed reserved windows, while traditional frame preemption mechanisms, though improving efficiency, still suffer from unpredictable TT frame delays. To address these issues, this paper proposes and validates two novel frame preemption strategies. First, a Guard Band Compression and Non-Disruptive Frame Preemption model (GCNFP) is introduced, which compresses the guard band to 76 bytes and employs a padding mechanism to ensure zero-offset handling of ET frames within the guard window, thus eliminating transmission deviations caused by ET blocking and inter-frame gaps (IFG). Second, an enhanced model with judgment mechanism (JM-GCNFP) is developed, which dynamically monitors ET frame transmission within the guard window and delays preemption until the optimal moment. This adaptive strategy maximizes the use of available time slots while reducing unnecessary preemption and padding overhead. Simulation results show that both strategies achieve zero-offset scheduling and significantly improve system performance. Specifically, JM-GCNFP reduces the maximum ET frame delay by more than 40% and improves bandwidth utilization by nearly 20% under high-load scenarios, demonstrating the advantages in achieving enhanced network determinism and transmission efficiency.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103697"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145941385","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":"GenMClass: Design and comparative analysis of genome classifier-on-chip platform","authors":"Daria Bromot ,&nbsp;Yehuda Kra ,&nbsp;Zuher Jahshan ,&nbsp;Esteban Garzón ,&nbsp;Adam Teman ,&nbsp;Leonid Yavits","doi":"10.1016/j.sysarc.2026.103702","DOIUrl":"10.1016/j.sysarc.2026.103702","url":null,"abstract":"<div><div>We propose GenMClass, a genome classification system-on-chip (SoC) implementing two different classification approaches and comprising two separate classification engines: a DNN accelerator GenDNN, that classifies DNA reads converted to images using a classification neural network, and a similarity search-capable Error Tolerant Content Addressable Memory ETCAM, that classifies genomes by k-mer matching. Classification operations are controlled by an embedded RISCV processor. GenMClass classification platform was designed and manufactured in a commercial 65 nm process. We conduct a comparative analysis of ETCAM and GenDNN classification efficiency as well as their performance, silicon area and power consumption using silicon measurements. The size of GenMClass SoC is 3.4 mm<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span> and its total power consumption (assuming both GenDNN and ETCAM perform classification at the same time) is 144 mW. This allows using GenMClass as a portable classifier for pathogen surveillance during pandemics, food safety and environmental monitoring, agriculture pathogen and antimicrobial resistance control, in the field or at points of care.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103702"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981625","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 survey on data management for Out-of-Core GNN systems","authors":"Junchi Ren ,&nbsp;Chao Li ,&nbsp;Xiaowei Chen ,&nbsp;Yao Chen ,&nbsp;Zehao Chen ,&nbsp;Qian Wei ,&nbsp;Zhaoyan Shen","doi":"10.1016/j.sysarc.2026.103684","DOIUrl":"10.1016/j.sysarc.2026.103684","url":null,"abstract":"<div><div>Graph neural networks (GNNs) have emerged as a powerful model for their effectiveness in learning over graphs, with broad applications in domains such as biology, e-commerce, and materials science. With the rapid growth of real-world graphs, efficient data management in GNNs has become a formidable challenge. Out-of-core (OOC) GNN system, as a representative solution, leverages external storage (CPU memory and SSD) to enable large-scale graph training on a single machine. Based on the scale of graph data, OOC GNN systems adopt different levels of storage extension and can be classified into two categories, semi OOC and fully OOC systems. However, the optimization details for both categories of OOC GNN systems remain only preliminarily understood. To address this gap, we provide a comprehensive survey of existing optimization techniques for semi OOC and fully OOC systems from the perspective of data management. We decompose the data management mechanisms into three layers: data storage, data organization, and data transfer, where data storage refers to the placement of graph data on disk, data organization pertains to the adaptive strategy that decides where to place graph data across different memory hierarchies during training, and data transfer concerns the I/O path of data movement between these storage layers. For each layer, we discuss the key challenges, review the corresponding optimization strategies proposed in existing OOC GNN systems, and analyze their advantages and limitations. Furthermore, we outline future research directions for data management of OOC GNN systems.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103684"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145981640","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":"Privacy-preserving access control and trust management for multi-authority in IoMT systems","authors":"Chenlu Xie,&nbsp;Xiaolin Gui","doi":"10.1016/j.sysarc.2026.103719","DOIUrl":"10.1016/j.sysarc.2026.103719","url":null,"abstract":"<div><div>Internet of Medical Things (IoMT) devices generate a huge amount of real-time data on a daily basis, which can be analyzed by medical practitioners to optimize diagnosis and treatment. Due to the complexity of devices and users in IoMT systems, robust measures are needed to ensure security and quality of service or information. However, most existing schemes require a trusted central authority to generate secret keys for users, which is often impractical in real-world scenarios. Although many multi-authority access control schemes have been proposed to address this issue, they still lack stronger defense and supervision mechanisms to effectively regulate user access. In this paper, we propose a privacy-preserving multi-authority access control scheme that enables policy hiding and efficiently prevents malicious access attacks. Specifically, multiple untrusted authorities independently generate attribute keys through secure two-party computation and zero-knowledge proofs. Even if multiple authorities collude, they cannot trace secret keys. Furthermore, the scheme enhances privacy by breaking the mapping between attributes and the access matrix. What is more, we also construct a dynamic access control mechanism based on trust management, which can effectively curb persistent access attacks by malicious data users. Our security analysis and experimental results show that the scheme achieves semantic security, resists collusion attacks, and constrains the malicious behavior of data users with minimal online encryption computational costs compared to other schemes.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103719"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146090487","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":"AIMD: AI-powered android malware detection for securing AIoT devices and networks using graph embedding and ensemble learning","authors":"Santosh K. Smmarwar ,&nbsp;Rahul Priyadarshi ,&nbsp;Pratik Angaitkar ,&nbsp;Subodh Mishra ,&nbsp;Rajkumar Singh Rathore","doi":"10.1016/j.sysarc.2026.103707","DOIUrl":"10.1016/j.sysarc.2026.103707","url":null,"abstract":"<div><div>The rapid evolution of Artificial Intelligence of Things (AIoT) is accelerating the development of smart societies, where interconnected consumer electronics such as smartphones, IoT devices, smart meters, and surveillance systems play a crucial role in optimizing operational efficiency and service delivery. However, this hyper-connected digital ecosystem is increasingly vulnerable to sophisticated Android malware attacks that exploit system weaknesses, disrupt services, and compromise data privacy and integrity. These malware variants leverage advanced evasion techniques, including permission abuse, dynamic runtime manipulation, and memory-based obfuscation, rendering traditional detection methods ineffective. The key challenges in securing AIoT-driven smart societies include managing high-dimensional feature spaces, detecting dynamically evolving malware behaviours, and ensuring real-time classification performance. To address these issues, this paper proposed an AI-powered Android Malware Detection (AIMD) framework designed for AIoT-enabled smart society environments. The framework extracts multi-level features (permissions, intents, API calls, and obfuscated memory patterns) from Android APK files and employs graph embedding techniques (DeepWalk and Node2Vec) for dimensionality reduction. Feature selection is optimized using the Red Deer Algorithm (RDA), a metaheuristic approach, while classification is performed through an ensemble of machine learning models (Support Vector Machine, Decision Tree, Random Forest, Extra Trees) enhanced by bagging, boosting, stacking, and soft voting techniques. Experimental evaluations on CICInvesAndMal2019 and CICMalMem2022 datasets demonstrate the effectiveness of the proposed system, achieving malware detection accuracies of 98.78% and 99.99%, respectively. By integrating AI-driven malware detection into AIoT infrastructures, this research advances cybersecurity resilience, safeguarding smart societies against emerging threats in an increasingly connected world.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103707"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039168","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":"GAS: A scheduling primitive dependency analysis-based cost model for tensor program optimization","authors":"Yonghua Hu ,&nbsp;Anxing Xie ,&nbsp;Yaohua Wang ,&nbsp;Zhe Li ,&nbsp;Zenghua Cheng ,&nbsp;Junyang Tang","doi":"10.1016/j.sysarc.2026.103721","DOIUrl":"10.1016/j.sysarc.2026.103721","url":null,"abstract":"<div><div>Automatically generating high-performance tensor programs has become a promising approach for deploying deep neural networks. A key challenge lies in designing an effective cost model to navigate the vast scheduling search space. Existing approaches typically fall into two categories, each with limitations: offline learning cost models rely on large pre-collected datasets, which may be incomplete or device-specific, and online learning cost models depend on handcrafted features, requiring substantial manual effort and expertise.</div><div>We propose GAS, a lightweight framework for generating tensor programs for deep learning applications. GAS reformulates feature extraction as a sequence-dependent analysis of scheduling primitives. Our cost model integrates three key factors to uncover performance-critical insights within scheduling sequences: (1) decision factors allocation, quantifying entropy and skewness of scheduling primitive factors to capture their dominance; (2) primitive contribution weights, measuring the relative impact of primitives on overall performance; and (3) structural semantic alignment, capturing correlations between scheduling primitive factors and hardware parallelism mechanisms. This approach reduces the complexity of handcrafted feature engineering and extensive pre-training datasets, significantly improving both efficiency and scalability. Experimental results on NVIDIA GPUs demonstrate that GAS achieves average speedups of 3.79<span><math><mo>×</mo></math></span> over AMOS and 2.22<span><math><mo>×</mo></math></span> over Ansor, while also consistently outperforming other state-of-the-art tensor compilers.</div></div>","PeriodicalId":50027,"journal":{"name":"Journal of Systems Architecture","volume":"173 ","pages":"Article 103721"},"PeriodicalIF":4.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039223","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}