IEEE Open Journal of the Computer Society最新文献

{"title":"GEN Self-Labeling Object Detector for PCB Recycling Evaluation","authors":"Leandro Honorato de S. Silva;Agostinho Freire;George O. A. Azevedo;Sérgio Campello Oliveira;Carlo M. R. da Silva;Bruno J. T. Fernandes","doi":"10.1109/OJCS.2025.3584297","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3584297","url":null,"abstract":"Waste Printed Circuit Boards (WPCBs) contain many valuable and rare metals found in electronic waste, and recycling these boards can help recover these metals and prevent hazardous elements from harming the environment. However, the diverse composition of PCBs makes it challenging to automate the recycling process, which should ideally be tailored to each PCB’s composition. Computer vision is a possible solution to evaluate WPCBs, but most state-of-the-art models depend on labeled datasets unavailable in the WPCB domain. Building a large and fully labeled WPCB dataset is expensive and time-consuming. In addition, the presence of long-tailed class imbalance, where specific electronic components are significantly more prevalent than others, further complicates the development of accurate detection and classification models. To address this, we propose a new method called GEN Self-Labeling Electronic Component Detector, which utilizes a domain adaptation strategy to train semi-supervised teacher-student models that can handle the lack of fully labeled datasets while mitigating the effects of class imbalance. We also introduce a new version of the Waste Printed Circuit Board Economic Feasibility Assessment (WPCB-EFAv2), which characterizes the PCB’s composition by identifying hazardous components, calculating the density of each component type, and estimating the metals that could be recovered from recycling electrolytic capacitors and integrated circuits. Finally, we present a case study involving six PCBs with different characteristics, from which we estimated that 121 g of metals could be recovered. The most recovered metal (108 g) was aluminum from electrolytic capacitors. This information can help reduce the PCB’s composition uncertainty, leading to more efficient dismantling and cost-effective recycling processes.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1041-1052"},"PeriodicalIF":0.0,"publicationDate":"2025-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11058390","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144634817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid Approach for WDM Network Restoration: Deep Reinforcement Learning and Graph Neural Networks","authors":"Isaac Ampratwum;Amiya Nayak","doi":"10.1109/OJCS.2025.3583945","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3583945","url":null,"abstract":"Ensuring robust and efficient service restoration in Wavelength Division Multiplexing (WDM) networks is crucial for maintaining network reliability amidst failures caused by disasters, equipment malfunctions, or power outages. This article presents a hybrid framework that integrates Deep Reinforcement Learning (DRL) and Graph Neural Networks (GNN) to optimize WDM network restoration. The proposed method leverages the decision-making capabilities of DRL and the graph-structured learning potential of GNN to dynamically adapt to network disruptions. By modeling network topology as a graph, the GNN extracts structural features, while the DRL agent intelligently selects restoration paths, balancing network load and minimizing restoration time. Experimental evaluations across multiple network topologies and failure scenarios reveal that the hybrid DRL+GNN approach outperforms conventional restoration techniques in terms of restoration success rate, resource utilization, and scalability. The framework’s ability to generalize across diverse network configurations highlights its adaptability and potential for deployment in real-world optical communication systems. This study underscores the transformative impact of combining AI techniques on advancing WDM network resilience and recovery capabilities.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1012-1026"},"PeriodicalIF":0.0,"publicationDate":"2025-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11054280","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144606288","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Graph Split Federated Learning for Distributed Large-Scale AIoT in Smart Cities","authors":"Hanyue Xu;Kah Phooi Seng;Li-Minn Ang;Wei Wang;Jeremy Smith","doi":"10.1109/OJCS.2025.3583271","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3583271","url":null,"abstract":"The rise of smart cities has leveraged the power of Internet of Things devices to transform urban services. A key element of this transformation is the widespread deployment of IoT devices for data collection, which feeds into machine learning algorithms to improve city services. However, the centralization of sensitive IoT data for ML raises privacy and efficiency concerns. Distributed collaborative machine learning, particularly split federated learning, has emerged as a solution, enabling privacy-preserving, resource-efficient training on IoT devices. This article introduces a novel SFL-based framework for graph convolutional neural networks, SFLGCN, which includes two variants SFLGCN (general) and SFLGCN-PP (Privacy Preservation), specifically designed for resource-constrained IoT systems in smart cities. SFLGCN-PP, an enhanced version of the framework, focuses on privacy preservation and is capable of handling graph-structured data, which is common in smart city scenarios, without requiring pre-defined adjacency matrices, thus enhancing data privacy. The framework’s efficacy is validated through predictive modeling of autonomous vehicle passenger demand using real-world IoT data. Additionally, the generalization capability of our framework is demonstrated on public graph datasets, where it outperforms traditional federated learning in graph neural network tasks, particularly in large-scale IoT environments with varying data distributions and client capacities.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1027-1040"},"PeriodicalIF":0.0,"publicationDate":"2025-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11050992","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144634647","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Pneumonia Diagnosis Through AI Interpretability: Comparative Analysis of Pixel-Level Interpretability and Grad-CAM on X-ray Imaging With VGG19","authors":"Mohammad Ennab;Hamid Mcheick","doi":"10.1109/OJCS.2025.3582726","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3582726","url":null,"abstract":"Pneumonia is a leading cause of morbidity and mortality worldwide, necessitating timely and precise diagnosis for effective treatment. Chest X-rays are the primary diagnostic tool, but their interpretation demands substantial expertise. Recent advancements in AI have shown promise in enhancing pneumonia detection from X-ray images, yet the opacity of deep learning models raises concerns about their clinical adoption. Interpretability in AI models is vital for fostering trust among healthcare professionals by providing transparency in decision-making processes. This study conducts a comparative analysis of two interpretability methods, Pixel Level Interpretability (PLI) and Gradient-weighted Class Activation Mapping (Grad-CAM), in the context of pneumonia classification using VGG19 on X-ray datasets. The research includes an experiment involving three distinct X-ray datasets. VGG19 is applied to classify a query image, and both PLI and Grad-CAM are used to interpret the classification decisions. The study evaluates these interpretability methods across multiple dimensions: computational efficiency, diagnostic performance, explanation continuity, calibration accuracy, robustness to training parameters, and feedback from medical experts. Our findings aim to determine which interpretability technique offers a more clinically meaningful explanation, balancing computational feasibility and diagnostic reliability. This study contributes to the development of explainable AI in healthcare, supporting the integration of trustworthy AI systems in clinical environments for enhanced pneumonia diagnosis.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1155-1165"},"PeriodicalIF":0.0,"publicationDate":"2025-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11049939","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144716205","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid Contrastive Learning With Attention-Based Neural Networks for Robust Fraud Detection in Digital Payment Systems","authors":"Md Shahin Alam Mozumder;Mohammad Balayet Hossain Sakil;Md Rokibul Hasan;Md Amit Hasan;K. M Nafiur Rahman Fuad;M. F. Mridha;Md Rashedul Islam;Yutaka Watanobe","doi":"10.1109/OJCS.2025.3581950","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3581950","url":null,"abstract":"Fraud detection in digital payment systems is a critical challenge due to the growing complexity of transaction patterns and the inherent class imbalance in datasets. This article proposes a novel Hybrid Contrastive Learning framework integrating Siamese Networks with Attention-Based Neural Networks to effectively distinguish fraudulent from legitimate transactions. The proposed model achieves state-of-the-art results, surpassing 10 recent methods in key metrics, with a recall of 95.42%, precision of 97.35%, and ROC-AUC of 98.78% on the Credit Card Fraud Detection dataset. Cross-dataset evaluations using a simulated transaction dataset demonstrate consistent generalization, achieving a recall of 95.12% and ROC-AUC of 98.60%. An ablation study underscores the impact of attention mechanisms and contrastive learning, with the combined approach improving F1-score by up to 2.64%. Additionally, SHAP-based analysis reveals the importance of key features such as transaction amount and PCA-derived components in model decisions, enhancing explainability. The results establish the proposed framework as a robust, interpretable, and scalable solution for fraud prevention in digital payment systems.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1053-1064"},"PeriodicalIF":0.0,"publicationDate":"2025-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11045880","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144640947","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"DSEM-NIDS: Enhanced Network Intrusion Detection System Using Deep Stacking Ensemble Model","authors":"Loreen Mahmoud;Madhusanka Liyanage;Jitin Singla;Sugata Gangopadhyay","doi":"10.1109/OJCS.2025.3581036","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3581036","url":null,"abstract":"The need to deploy a network intrusion detection system (NIDS) is essential and has become increasingly necessary for every network, regardless whether it is wired, wireless, or hybrid, and its purpose is commercial, medical, defense, or social. Since the amount of data transfer over the Internet increases every year, using a single model as an IDS to secure the network cannot be considered enough as it may have many problems like high bias or high variance, which lead to high rates of false negatives and false positives. In this article, we propose an ensemble learning-based NIDS (DSEM-NIDS); this system is a deep-stacking model with a nested structure that has the ability to score a high performance with low false positive and low false negative rates. Four datasets are used as a benchmark to evaluate the proposed model: The 5G-NIDD, UNR-IDD, N-BaIoT, and NSL-KDD datasets. The results show that the proposed deep stacking model is robust, has good scalability, has the ability to distinguish between classes, and has the flexibility to adapt to different input data. It also performs better than other used models.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"955-967"},"PeriodicalIF":0.0,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11045004","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144598035","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fast and Private 1-to-$N$ Face Identification Protocols","authors":"Nuttapong Attrapadung;Reo Eriguchi;Goichiro Hanaoka;Takahiro Matsuda;Naohisa Nishida;Tatsumi Oba;Jacob C. N. Schuldt;Koki Tejima;Tadanori Teruya;Yuji Unagami;Naoto Yanai","doi":"10.1109/OJCS.2025.3580739","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3580739","url":null,"abstract":"Face identification is a pivotal aspect of user authentication across diverse domains, spanning from smartphone security to access control in high-security environments. Privacy-preserving face identification aims to authenticate users using facial images while preserving their privacy. In this article, we focus on privacy-preserving protocols for one-to-many (1-to-<inline-formula><tex-math>$N$</tex-math></inline-formula>) face identification. Such protocols enable the authentication of individuals from a pool of registered users without disclosing their identity among the group. We present new protocols based on secret-sharing based secure multi-party computation (MPC). Our initial (warm-up) protocol directly applies MPC to the plain ArcFace framework, marking the first instance of such a face identification scheme without reliance on homomorphic encryption, a primary tool in previous works. Notably, this protocol exhibits efficiency for small-scale databases, requiring approximately 1 second for authentication among 1000 users. Building upon this foundation, our main contribution lies in our second protocol, designed to enhance scalability via a new approach to operations on large-scale databases. It significantly improves runtime performance compared to the state-of-the-art scheme of Bai et al., achieving approximately 2.31 times, 4.59 times, and 6.80 times faster authentication for registered user databases of sizes <inline-formula><tex-math>$N{=}10{,}000$</tex-math></inline-formula> and <inline-formula><tex-math>$N{=}100{,}000$</tex-math></inline-formula>, and <inline-formula><tex-math>$N{=}1{,}000{,}000$</tex-math></inline-formula>, respectively. Notably, our protocol enables user authentication in about a second for the first time in the case of database with 30,000 users. While our second protocol offers fast authentication times, it does entail some leakage of intermediate values. Nevertheless, this leakage is minimal and far less than that of previous works that allow leakage. Through our contributions, we aim to propel the state-of-the-art in face identification protocols, striking a balance between the imperatives of efficiency and privacy in real-world applications.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"968-986"},"PeriodicalIF":0.0,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11045323","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144597651","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Radio Frequency Sensing–Based Human Emotion Identification by Leveraging 2D Transformation Techniques and Deep Learning Models","authors":"Najah AbuAli;Ihtesham Jadoon;Farman Ullah;Mohammad Hayajneh;Shayma Alkobaisi","doi":"10.1109/OJCS.2025.3580570","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3580570","url":null,"abstract":"To meet the need for a reliable, contactless, and noninvasive sensing platform for emotion recognition, The one-dimensional (1D) time-series signal dataset obtained using the designed RFS–SDR platform is processed and transformed into two-dimensional (2D) time–frequency images (TFIs) via continuous wavelet transform (CWT), short-time Fourier transform (STFT), and wavelet coherence transform (WCT). These TFIs are then fed into pretrained deep learning models—AlexNet, ResNet18, and GoogleNet—to extract features for identifying eight emotions using varying batch sizes and optimizers. The core objective is to evaluate the effectiveness of deep learning models from transformed time-frequency features, comparing their performance with different transformation techniques and training conditions. The results show that AlexNet consistently outperforms other models, achieving superior accuracy, precision, recall, and F1 scores of up to 98%, particularly when combined with the SGDM optimizer and CWT features. ResNet18 shows superior performance, with accuracy reaching 99% when paired with Adam optimizer and CWT features; furthermore, GoogleNet exhibits high accuracy with Adam. AlexNet is robust and maintains high precision and recall across all configurations. Computational analysis reveals that AlexNet is time-efficient, particularly at large batch sizes, while GoogleNet incurs higher computational costs due to its complex architecture. The study underscores the impacts of the optimizer selection, batch size, and feature extraction methods on model performance and computational efficiency, offering valuable insights for optimizing deep learning models for RFS-based human emotion recognition.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1178-1189"},"PeriodicalIF":0.0,"publicationDate":"2025-06-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11039034","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144716204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"LLMs on a Budget: System-Level Approaches to Power-Efficient and Scalable Fine-Tuning","authors":"Kailash Gogineni;Ali Suvizi;Guru Venkataramani","doi":"10.1109/OJCS.2025.3580498","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3580498","url":null,"abstract":"Large Language Models (LLMs) have shown remarkable capabilities in various applications, including robotics, telecommunications, and scientific discovery. While much attention has been given to LLM inference and training phases, fine-tuning has received less focus despite its increasing cost, especially from a systems perspective. Fine-tuning is especially important for customizing compact models for edge applications, such as personal assistants running on local devices and models personalized with user-specific data, which in turn requires a deeper examination of fine-tuning performance and efficiency on single-GPU systems. Fine-tuning large models involves intensive matrix operations from backpropagation and gradient updates, which require extensive power and memory usage. In order to explore the range of performance optimization opportunities available to improve the LLM fine-tuning runtime, we understand the impact of techniques like activation checkpointing, low-rank adaptation, and operation fusion on LLM fine-tuning runtime optimization. In addition, we explore the effects of resource utilization through GPU peak power capping. Our experiments, conducted on NVIDIA RTX 4090 GPU using Meta’s LLaMA-3.1, Google’s Gemma, and Microsoft’s Phi-3, reveal that enabling all optimizations reduces memory usage by over 40% compared to FP32 baselines. Moreover, power capping to 300 W results in an average throughput drop of only 5.55% while reducing power consumption by 33%. Post-fine-tuning accuracy improvements on the Sycophancy Evaluation Benchmark range from 2% to 5%, depending on model architecture, validating that our optimization techniques preserve model quality while reducing resource requirements. Furthermore, we discuss several insights and potential future research directions from a systems perspective.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"987-1000"},"PeriodicalIF":0.0,"publicationDate":"2025-06-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11037824","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144598036","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Conv-Ensemble for Solar Power Prediction With First Nations Seasonal Information","authors":"Selvarajah Thuseethan;Sandipkumar Gangajaliya;Luke Hamlin;Bharanidharan Shanmugam;Suresh Thennadil","doi":"10.1109/OJCS.2025.3580339","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3580339","url":null,"abstract":"Power generation forecasting, especially for solar power, is crucial for future energy planning. In this study, a novel framework, namely <italic>FNS-Metrics</i>, is proposed to integrate seasonal information from First Nations calendars into solar power forecasting. Furthermore, a novel Conv-Ensemble framework is proposed, leveraging the high-level feature extraction capabilities of Conv1D layers along with the low-level feature extraction abilities of transformer and LSTM networks. A weighted feature concatenation technique is also integrated into the proposed approach to combine the features effectively. To validate the proposed FNS-Metrics and Conv-Ensemble framework, a new dataset is constructed by collecting power and weather data from the Desert Knowledge Australia Solar Center in Alice Springs and integrating data related to First Nations seasonal cycles. Experiments on this dataset show that the Conv-Ensemble framework with FNS-Metrics outperforms traditional approaches, achieving state-of-the-art solar power prediction with the highest <inline-formula><tex-math>$R^{2}$</tex-math></inline-formula> of 0.8641 and the lowest MSE of 22.41. These represent a 14.60% and 26.21% increase compared to the baseline configuration of Conv-Transformer. The ablation study demonstrates that the Conv-Ensemble framework improves performance compared to the baselines. Furthermore, the results for individual and combined FNS-Metrics features show a progressive improvement in performance.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"884-895"},"PeriodicalIF":0.0,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11037463","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144519400","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}