Pub Date : 2025-09-16DOI: 10.1109/OJCS.2025.3610270
Eran Dahan;Yosi Keller
In this article, we aim to improve kinship verification performance by optimizing embedding representations tailored to each kinship relation type. We concentrate on two relationship categories: same-generation (e.g., Brothers, Sisters, Siblings) and mixed-generation (e.g., Father-Daughter, Mother-Son). For mixed-generation relationships, we develop a sophisticated contrastive learning framework that takes advantage of the hierarchical structure within a family, such as refining the kinship relation embedding for Mother-Daughter as an extension to the Sisters relationship. For the types of same-generation relationships, we propose a tailored contrastive learning scheme for each specific kinship relationship. Further, we developed a unique sampling method for our scheme which helps to reduce the overfitting of the kinship verification task. Overall, our method achieves state-of-the-art performance on the FIW dataset, outperforming previous benchmarks by a substantial margin.
{"title":"Kinship Verification Using Hierarchical Structures and Extended Contrastive Learning","authors":"Eran Dahan;Yosi Keller","doi":"10.1109/OJCS.2025.3610270","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3610270","url":null,"abstract":"In this article, we aim to improve kinship verification performance by optimizing embedding representations tailored to each kinship relation type. We concentrate on two relationship categories: same-generation (e.g., <italic>Brothers</i>, <italic>Sisters</i>, <italic>Siblings</i>) and mixed-generation (e.g., <italic>Father-Daughter</i>, <italic>Mother-Son</i>). For mixed-generation relationships, we develop a sophisticated contrastive learning framework that takes advantage of the hierarchical structure within a family, such as refining the kinship relation embedding for <italic>Mother-Daughter</i> as an extension to the <italic>Sisters</i> relationship. For the types of same-generation relationships, we propose a tailored contrastive learning scheme for each specific kinship relationship. Further, we developed a unique sampling method for our scheme which helps to reduce the overfitting of the kinship verification task. Overall, our method achieves state-of-the-art performance on the FIW dataset, outperforming previous benchmarks by a substantial margin.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1549-1560"},"PeriodicalIF":0.0,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11164899","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145351889","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}
Pub Date : 2025-09-11DOI: 10.1109/OJCS.2025.3609195
Pascal A. Schirmer;Dimitrios Kolosov;Iosif Mporas
Non-Intrusive Load Monitoring (NILM) aims to estimate the power consumption of electrical appliances from the aggregated power consumption. While recent machine learning approaches have demonstrated very high disaggregation accuracies, ensuring real-time capability is crucial in NILM’s hardware implementations. We propose a constrained elastic matching approach for NILM to reduce execution time significantly. Our approach was tested on two datasets (REDD and AMPds2). The reported performance is on average 93.2% in terms of estimation accuracy for deferrable loads using the AMPds2 dataset. The proposed approach reduces execution time by a factor of ten compared to unconstrained elastic matching techniques, achieving per-frame inference times of 3.5–12.1 ms depending on the hardware platform and model size. Memory usage for the largest model is approximately 7.5 MB, and reducing the model to 10% of reference signatures lowers active power consumption from 12.1 W to 5.2 W, representing a 57% energy saving with minimal accuracy loss. Furthermore, the proposed approach has been evaluated on five different microprocessors, demonstrating consistent runtime reduction and enabling real-time implementation of elastic matching based NILM with large reference databases.
{"title":"Multivariate Constrained Elastic Matching With Application in Real-Time Energy Disaggregation","authors":"Pascal A. Schirmer;Dimitrios Kolosov;Iosif Mporas","doi":"10.1109/OJCS.2025.3609195","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3609195","url":null,"abstract":"Non-Intrusive Load Monitoring (NILM) aims to estimate the power consumption of electrical appliances from the aggregated power consumption. While recent machine learning approaches have demonstrated very high disaggregation accuracies, ensuring real-time capability is crucial in NILM’s hardware implementations. We propose a constrained elastic matching approach for NILM to reduce execution time significantly. Our approach was tested on two datasets (REDD and AMPds2). The reported performance is on average 93.2% in terms of estimation accuracy for deferrable loads using the AMPds2 dataset. The proposed approach reduces execution time by a factor of ten compared to unconstrained elastic matching techniques, achieving per-frame inference times of 3.5–12.1 ms depending on the hardware platform and model size. Memory usage for the largest model is approximately 7.5 MB, and reducing the model to 10% of reference signatures lowers active power consumption from 12.1 W to 5.2 W, representing a 57% energy saving with minimal accuracy loss. Furthermore, the proposed approach has been evaluated on five different microprocessors, demonstrating consistent runtime reduction and enabling real-time implementation of elastic matching based NILM with large reference databases.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1475-1487"},"PeriodicalIF":0.0,"publicationDate":"2025-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11159158","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141625","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}
The Industrial Internet of Things (IIoT) generates a vast volume of sensitive data that demands not only confidentiality but also authenticity and integrity—especially in large-scale deployments. Ensuring that data originates from trusted devices is critical; however, existing authentication mechanisms often lack scalability and effective revocation support. To address these challenges, we propose LightPUF-IIoT, a secure and lightweight authentication scheme designed for fog-assisted IIoT data sharing. The scheme leverages Physical Unclonable Functions (PUFs) and Non-Interactive Zero-Knowledge Proofs (NIZKPs) to enable scalable, group-based authentication for devices and fog nodes. By binding authenticated identities to cryptographic tokens used during data transmission, LightPUF-IIoT ensures data authenticity and supports real-time rogue device detection. The scheme also includes efficient mechanisms for device revocation and secure token regeneration. Experimental results show that LightPUF-IIoT provides strong security guarantees with minimal resource overhead and significantly outperforms existing approaches in terms of computational cost, scalability, and authentication throughput.
{"title":"LightPUF-IIoT: A Lightweight PUF-Based Authentication Scheme With Real-Time Detection of Rogue Devices in Fog-Assisted IIoT Data Sharing","authors":"Somchart Fugkeaw;Archawit Changtor;Thanabordee Maneerat;Pakapon Rattanasrisuk;Kittipat Tangtanawirut","doi":"10.1109/OJCS.2025.3607984","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3607984","url":null,"abstract":"The Industrial Internet of Things (IIoT) generates a vast volume of sensitive data that demands not only confidentiality but also authenticity and integrity—especially in large-scale deployments. Ensuring that data originates from trusted devices is critical; however, existing authentication mechanisms often lack scalability and effective revocation support. To address these challenges, we propose <bold>LightPUF-IIoT</b>, a secure and lightweight authentication scheme designed for fog-assisted IIoT data sharing. The scheme leverages <bold>Physical Unclonable Functions (PUFs)</b> and <bold>Non-Interactive Zero-Knowledge Proofs (NIZKPs)</b> to enable scalable, group-based authentication for devices and fog nodes. By binding authenticated identities to cryptographic tokens used during data transmission, LightPUF-IIoT ensures data authenticity and supports real-time rogue device detection. The scheme also includes efficient mechanisms for device revocation and secure token regeneration. Experimental results show that LightPUF-IIoT provides strong security guarantees with minimal resource overhead and significantly outperforms existing approaches in terms of computational cost, scalability, and authentication throughput.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1438-1450"},"PeriodicalIF":0.0,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11153801","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141632","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}
Pub Date : 2025-09-09DOI: 10.1109/OJCS.2025.3607975
Suchuan Xing;Yihan Wang
Anomaly detection in system software traditionally relies on single-modal algorithms that analyze either discrete log events or continuous performance metrics in isolation, potentially missing complex anomalies that manifest across both modalities. We present a novel deep learning framework that leverages cross-modal attention mechanisms to jointly model log sequences and performance metrics for enhanced anomaly detection. Our method proposes Long Short-Term Memory (LSTM) networks to capture temporal dependencies in log event sequences and Temporal Convolutional Networks (TCNs) to model performance metric time series. The core innovation lies in our Cross-Modal Attention Mechanism, which dynamically weighs log events and metric features based on inter-modal relationships, enabling the detection of subtle anomalies that require contextual information from both data sources. Unlike conventional multi-modal fusion techniques that merely concatenate features, our attention mechanism explicitly models the dependencies between log patterns and metric behaviors, allowing the network to focus on relevant log events during metric anomalies and vice versa. We conduct comprehensive experiments on public datasets including HDFS and BGL logs paired with cloud computing performance metrics, as well as real-world cloud environments. Our method achieves significant improvements over single-modal baselines, with F1-scores increasing by 12.3% on average across datasets. Ablation studies confirm the effectiveness of the cross-modal attention mechanism, while real-time deployment experiments using Apache Flink demonstrate practical applicability with sub-second latency. The proposed framework addresses a critical gap in system software monitoring by providing a principled approach to multi-modal anomaly detection that scales to enterprise-level deployments.
{"title":"Cross-Modal Attention Networks for Multi-Modal Anomaly Detection in System Software","authors":"Suchuan Xing;Yihan Wang","doi":"10.1109/OJCS.2025.3607975","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3607975","url":null,"abstract":"Anomaly detection in system software traditionally relies on single-modal algorithms that analyze either discrete log events or continuous performance metrics in isolation, potentially missing complex anomalies that manifest across both modalities. We present a novel deep learning framework that leverages cross-modal attention mechanisms to jointly model log sequences and performance metrics for enhanced anomaly detection. Our method proposes Long Short-Term Memory (LSTM) networks to capture temporal dependencies in log event sequences and Temporal Convolutional Networks (TCNs) to model performance metric time series. The core innovation lies in our Cross-Modal Attention Mechanism, which dynamically weighs log events and metric features based on inter-modal relationships, enabling the detection of subtle anomalies that require contextual information from both data sources. Unlike conventional multi-modal fusion techniques that merely concatenate features, our attention mechanism explicitly models the dependencies between log patterns and metric behaviors, allowing the network to focus on relevant log events during metric anomalies and vice versa. We conduct comprehensive experiments on public datasets including HDFS and BGL logs paired with cloud computing performance metrics, as well as real-world cloud environments. Our method achieves significant improvements over single-modal baselines, with F1-scores increasing by 12.3% on average across datasets. Ablation studies confirm the effectiveness of the cross-modal attention mechanism, while real-time deployment experiments using Apache Flink demonstrate practical applicability with sub-second latency. The proposed framework addresses a critical gap in system software monitoring by providing a principled approach to multi-modal anomaly detection that scales to enterprise-level deployments.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1463-1474"},"PeriodicalIF":0.0,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11153984","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141630","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}
Pub Date : 2025-08-28DOI: 10.1109/OJCS.2025.3603234
Muhammad Iqbal;Tabinda Ashraf;Jen-Yi Pan
Reconfigurable Intelligent Surfaces (RIS) have emerged as a transformative solution for enhancing spectral and energy efficiency in wireless networks. However, conventional RIS architectures, whether passive or active, face significant limitations due to trade-offs involving power consumption, signal amplification, and deployment complexity. This article aims to overcome these limitations by developing an adaptive RIS architecture suitable for diverse transmission conditions. We propose a novel three-stage hybrid RIS system that dynamically switches among active, passive, and dormant modes based on channel quality and transmit power thresholds. A joint optimization framework is developed to enable adaptive mode selection, beamforming, and RIS configuration. This framework integrates mode-aware control logic and fractional programming to maximize system-wide sum-rate performance while minimizing energy consumption. Extensive simulations across varying propagation scenarios confirm that the proposed hybrid RIS outperforms conventional RIS designs in both spectral and energy efficiency. The results show that active mode yields high gains in low-power or obstructed channels, passive mode supports energy-efficient reflection under moderate conditions, and the dormant mode effectively conserves energy in high-power environments. Overall, the three-stage hybrid RIS architecture provides a robust, flexible, and high-performance solution, making it a promising candidate for future 6G wireless systems.
{"title":"Adaptive Three-Stage Hybrid RIS for Energy-Efficient and High-Performance Wireless Networks","authors":"Muhammad Iqbal;Tabinda Ashraf;Jen-Yi Pan","doi":"10.1109/OJCS.2025.3603234","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3603234","url":null,"abstract":"Reconfigurable Intelligent Surfaces (RIS) have emerged as a transformative solution for enhancing spectral and energy efficiency in wireless networks. However, conventional RIS architectures, whether passive or active, face significant limitations due to trade-offs involving power consumption, signal amplification, and deployment complexity. This article aims to overcome these limitations by developing an adaptive RIS architecture suitable for diverse transmission conditions. We propose a novel three-stage hybrid RIS system that dynamically switches among active, passive, and dormant modes based on channel quality and transmit power thresholds. A joint optimization framework is developed to enable adaptive mode selection, beamforming, and RIS configuration. This framework integrates mode-aware control logic and fractional programming to maximize system-wide sum-rate performance while minimizing energy consumption. Extensive simulations across varying propagation scenarios confirm that the proposed hybrid RIS outperforms conventional RIS designs in both spectral and energy efficiency. The results show that active mode yields high gains in low-power or obstructed channels, passive mode supports energy-efficient reflection under moderate conditions, and the dormant mode effectively conserves energy in high-power environments. Overall, the three-stage hybrid RIS architecture provides a robust, flexible, and high-performance solution, making it a promising candidate for future 6G wireless systems.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1451-1462"},"PeriodicalIF":0.0,"publicationDate":"2025-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11142651","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141633","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}
High-dimensional similarity search remains a critical challenge in machine learning, particularly when data lie on complex, non-linear manifolds that undermine the effectiveness of classical Locality-Sensitive Hashing (LSH). This work introduces Gaussian LSH, a kernel-based hashing framework that integrates over-clustering with Gaussian probability density modelling to improve locality preservation while maintaining computational efficiency. The method generates compact binary codes from a hybrid kernel–PDF score and supports scalable GPU-accelerated indexing for large datasets. Empirical evaluations across multiple visual and textual benchmarks demonstrate consistent improvements in recall and query latency compared to representative LSH variants and approximate nearest neighbour libraries. Gaussian LSH achieves recall gains of up to $text{9},text{pp}$ and latency reductions of up to $4.3times$, with benefits sustained across a range of code lengths. These results highlight the approach’s scalability and accuracy, supporting its use in medium- to large-scale similarity retrieval tasks across diverse data domains.
{"title":"Gaussian Kernel-Based LSH for High-Dimensional Similarity Search","authors":"Masrat Rasool;Khelil Kassoul;Samir Brahim Belhaouari","doi":"10.1109/OJCS.2025.3602355","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3602355","url":null,"abstract":"High-dimensional similarity search remains a critical challenge in machine learning, particularly when data lie on complex, non-linear manifolds that undermine the effectiveness of classical Locality-Sensitive Hashing (LSH). This work introduces Gaussian LSH, a kernel-based hashing framework that integrates over-clustering with Gaussian probability density modelling to improve locality preservation while maintaining computational efficiency. The method generates compact binary codes from a hybrid kernel–PDF score and supports scalable GPU-accelerated indexing for large datasets. Empirical evaluations across multiple visual and textual benchmarks demonstrate consistent improvements in recall and query latency compared to representative LSH variants and approximate nearest neighbour libraries. Gaussian LSH achieves recall gains of up to <inline-formula><tex-math>$text{9},text{pp}$</tex-math></inline-formula> and latency reductions of up to <inline-formula><tex-math>$4.3times$</tex-math></inline-formula>, with benefits sustained across a range of code lengths. These results highlight the approach’s scalability and accuracy, supporting its use in medium- to large-scale similarity retrieval tasks across diverse data domains.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1402-1413"},"PeriodicalIF":0.0,"publicationDate":"2025-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11134766","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141631","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}
Pub Date : 2025-08-22DOI: 10.1109/OJCS.2025.3601668
Abu Sadat Mohammad Shaker;Md Tohidul Islam;A T M Omor Faruq;Hritika Barua;Uland Rozario;M. F. Mridha;Md. Jakir Hossen
Customer segmentation is essential for personalized marketing, customer retention, and strategic decision-making. Traditional segmentation methods, such as k-Means and Gaussian Mixture Models, rely on static features and fail to capture the evolving nature of customer behavior. Existing methods also struggle to account for temporal dynamics, limiting their effectiveness in fast-changing markets. This article proposes TACS-Net, a Temporal-Aware Customer Segmentation Network that dynamically models behavioral shifts using Temporal Convolutional Networks (TCN), Transformers, and a Recurrent Clustering Algorithm (RCA). TACS-Net adapts to changes in customer purchasing patterns over time, offering superior segmentation accuracy and stability. It integrates short- and long-term behavioral modeling, providing a robust, real-time framework for continuous customer profiling. Evaluation on two real-world datasets (CSD1 and CSD2) demonstrates that TACS-Net achieves a silhouette score of 0.55 on CSD1 and 0.54 on CSD2, outperforming traditional baselines. The model also shows higher temporal stability, with 84.3% and 83.7% of customers retaining their segment over one month in CSD1 and CSD2, respectively, compared to 72.1% and 74.0% with k-Means. Explainability analysis using SHAP reveals key factors driving segmentation, such as spending score, purchase frequency, and last purchase amount. While TACS-Net outperforms existing methods in clustering quality and stability, its higher computational cost calls for further optimization.
{"title":"TACS-Net: Temporal-Aware Customer Segmentation Network","authors":"Abu Sadat Mohammad Shaker;Md Tohidul Islam;A T M Omor Faruq;Hritika Barua;Uland Rozario;M. F. Mridha;Md. Jakir Hossen","doi":"10.1109/OJCS.2025.3601668","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3601668","url":null,"abstract":"Customer segmentation is essential for personalized marketing, customer retention, and strategic decision-making. Traditional segmentation methods, such as k-Means and Gaussian Mixture Models, rely on static features and fail to capture the evolving nature of customer behavior. Existing methods also struggle to account for temporal dynamics, limiting their effectiveness in fast-changing markets. This article proposes TACS-Net, a Temporal-Aware Customer Segmentation Network that dynamically models behavioral shifts using Temporal Convolutional Networks (TCN), Transformers, and a Recurrent Clustering Algorithm (RCA). TACS-Net adapts to changes in customer purchasing patterns over time, offering superior segmentation accuracy and stability. It integrates short- and long-term behavioral modeling, providing a robust, real-time framework for continuous customer profiling. Evaluation on two real-world datasets (CSD1 and CSD2) demonstrates that TACS-Net achieves a silhouette score of 0.55 on CSD1 and 0.54 on CSD2, outperforming traditional baselines. The model also shows higher temporal stability, with 84.3% and 83.7% of customers retaining their segment over one month in CSD1 and CSD2, respectively, compared to 72.1% and 74.0% with k-Means. Explainability analysis using SHAP reveals key factors driving segmentation, such as spending score, purchase frequency, and last purchase amount. While TACS-Net outperforms existing methods in clustering quality and stability, its higher computational cost calls for further optimization.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1426-1437"},"PeriodicalIF":0.0,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11134290","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141627","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}
Pub Date : 2025-08-20DOI: 10.1109/OJCS.2025.3600632
Harsh Mehta;Santosh Kumar Bharti;Nishant Doshi
Automatic text summarization has been a prominent research topic for over a decade, aiming to distill concise summaries from extensive textual documents. This study introduces a novel approach addressing the intricacies of morphologically rich Indo-Iranian languages. We propose a unique method that leverages linguistic formality to guide summary generation. Building on an existing formality formula designed for English, we adapt and extend it for the structural characteristics of Indo-Iranian languages, which follow the Subject-Object-Verb (SOV) order. Our refined formula demonstrates a 7.28% improvement in formality scores compared to informal texts, validated through statistical significance testing. To assess sentence formality, we use our custom formula alongside additional features such as Shannon entropy scores and numeric token presence, combining these into a comprehensive sentence evaluation metric. Using this framework, we generate extractive summaries of Gujarati texts. Comparative evaluations at 20% and 30% compression ratios reveal that our method outperforms existing baselines, with ROUGE-1 score improvements of 14.63% at 30% and 28.60% at 20% compression. For reproducibility and further exploration, all experimental data and source code are made publicly available.
{"title":"Extractive Text Summarization Using Formality of Language","authors":"Harsh Mehta;Santosh Kumar Bharti;Nishant Doshi","doi":"10.1109/OJCS.2025.3600632","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3600632","url":null,"abstract":"Automatic text summarization has been a prominent research topic for over a decade, aiming to distill concise summaries from extensive textual documents. This study introduces a novel approach addressing the intricacies of morphologically rich Indo-Iranian languages. We propose a unique method that leverages linguistic formality to guide summary generation. Building on an existing formality formula designed for English, we adapt and extend it for the structural characteristics of Indo-Iranian languages, which follow the Subject-Object-Verb (SOV) order. Our refined formula demonstrates a 7.28% improvement in formality scores compared to informal texts, validated through statistical significance testing. To assess sentence formality, we use our custom formula alongside additional features such as Shannon entropy scores and numeric token presence, combining these into a comprehensive sentence evaluation metric. Using this framework, we generate extractive summaries of Gujarati texts. Comparative evaluations at 20% and 30% compression ratios reveal that our method outperforms existing baselines, with ROUGE-1 score improvements of 14.63% at 30% and 28.60% at 20% compression. For reproducibility and further exploration, all experimental data and source code are made publicly available.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1414-1425"},"PeriodicalIF":0.0,"publicationDate":"2025-08-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11130639","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141626","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}
Pub Date : 2025-08-20DOI: 10.1109/OJCS.2025.3600916
Patrick McEnroe;Shen Wang;Madhusanka Liyanage
With the expanding use of unmanned aerial vehicles (UAVs) across various fields, efficient obstacle avoidance has become increasingly crucial. This UAV obstacle avoidance can be achieved through deep reinforcement learning (DRL) algorithms deployed directly on-device (i.e., at the edge). However, practical deployment is constrained by high training time and high inference latency. In this paper, we propose methods to improve DRL-based UAV obstacle avoidance efficiency through improving both training efficiency and inference latency. To reduce inference latency, we employ input dimension reduction, streamlining the state representation to enable faster decision-making. For training time reduction, we leverage transfer learning, allowing the obstacle avoidance models to rapidly adapt to new environments without starting from scratch. To show the generalizability of our methods, we applied them to a discrete action space dueling double deep Q-network (D3QN) model and a continuous action space soft actor critic (SAC) model. Inference results are evaluated on both an NVIDIA Jetson Nano edge device and a NVIDIA Jetson Orin Nano edge device and we propose a combined method called FERO which combines state space reduction, transfer learning, and conversion to TensorRT for optimum deployment on NVIDIA Jetson devices. For our individual methods and combined method, we demonstrate reductions in training and inference times with minimal compromise in obstacle avoidance performance.
随着无人机在各个领域的广泛应用，高效避障变得越来越重要。这种无人机避障可以通过直接部署在设备上（即在边缘）的深度强化学习（DRL）算法来实现。然而，实际部署受到高训练时间和高推理延迟的限制。本文提出了通过提高训练效率和推理延迟来提高基于drl的无人机避障效率的方法。为了减少推理延迟，我们采用输入降维，简化状态表示以实现更快的决策。为了减少训练时间，我们利用迁移学习，使避障模型能够快速适应新环境，而无需从头开始。为了证明我们的方法的泛化性，我们将它们应用于离散动作空间决斗双深度q网络（D3QN）模型和连续动作空间软行为批评家（SAC）模型。在NVIDIA Jetson Nano edge设备和NVIDIA Jetson Orin Nano edge设备上对推理结果进行了评估，并提出了一种称为FERO的组合方法，该方法将状态空间约简、迁移学习和转换到TensorRT相结合，以便在NVIDIA Jetson设备上进行最佳部署。对于我们的单独方法和组合方法，我们证明了在最小程度上损害避障性能的情况下减少了训练和推理时间。
{"title":"FERO: Efficient Deep Reinforcement Learning based UAV Obstacle Avoidance at the Edge","authors":"Patrick McEnroe;Shen Wang;Madhusanka Liyanage","doi":"10.1109/OJCS.2025.3600916","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3600916","url":null,"abstract":"With the expanding use of unmanned aerial vehicles (UAVs) across various fields, efficient obstacle avoidance has become increasingly crucial. This UAV obstacle avoidance can be achieved through deep reinforcement learning (DRL) algorithms deployed directly on-device (i.e., at the edge). However, practical deployment is constrained by high training time and high inference latency. In this paper, we propose methods to improve DRL-based UAV obstacle avoidance efficiency through improving both training efficiency and inference latency. To reduce inference latency, we employ input dimension reduction, streamlining the state representation to enable faster decision-making. For training time reduction, we leverage transfer learning, allowing the obstacle avoidance models to rapidly adapt to new environments without starting from scratch. To show the generalizability of our methods, we applied them to a discrete action space dueling double deep Q-network (D3QN) model and a continuous action space soft actor critic (SAC) model. Inference results are evaluated on both an NVIDIA Jetson Nano edge device and a NVIDIA Jetson Orin Nano edge device and we propose a combined method called FERO which combines state space reduction, transfer learning, and conversion to TensorRT for optimum deployment on NVIDIA Jetson devices. For our individual methods and combined method, we demonstrate reductions in training and inference times with minimal compromise in obstacle avoidance performance.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1378-1389"},"PeriodicalIF":0.0,"publicationDate":"2025-08-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11130910","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141628","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}
Pub Date : 2025-08-18DOI: 10.1109/OJCS.2025.3599786
Uday K. Chakraborty;Cezary Z. Janikow
Computational complexity analysis of an algorithm is an integral part of understanding and applying that algorithm. For stochastic, adaptive heuristics in non-convex optimization, however, complexity analysis is often difficult. This article derives, for the first time in the literature, the complexity of the semi-steady-state Jaya algorithm (which is a recently developed variant of the Jaya algorithm) without the unimodality assumption. The Jaya algorithm, and its improvement, the semi-steady-state Jaya, are among the newest metaheuristics in population-based, nature-inspired optimization methods. In black-box function optimization, stochastic models of evolutionary and non-evolutionary heuristics often study the search process as sampling from distributions that are difficult to estimate. Unimodal distributions used for this purpose are easy to analyze but are necessarily restrictive. In this article, we model multimodality using mixtures of unimodal densities. For multimodal mixtures of uniform densities and, separately, of exponential densities (with different location parameters for the mixture components), analytical expressions, many of them closed-form, are derived for (i) the expectation of the largest order statistic for samples from the mixture; (ii) asymptotics of the above expectation for the large-sample case; (iii) survival probability corresponding to the (asymptotic) expected value of the largest order statistic; and (iv) asymptotics of sums of survival probabilities. The above quantities are used in a stochastic model of the semi-steady-state Jaya algorithm, obtaining the (asymptotic) expectation of the number of updates of the best individual in a population of the algorithm, which in turn is used in the derivation of the computational complexity of the algorithm.
{"title":"Using Extreme Order Statistics of Multimodal Mixture Distributions for Complexity Analysis of Semi-Steady-State Jaya Algorithm","authors":"Uday K. Chakraborty;Cezary Z. Janikow","doi":"10.1109/OJCS.2025.3599786","DOIUrl":"https://doi.org/10.1109/OJCS.2025.3599786","url":null,"abstract":"Computational complexity analysis of an algorithm is an integral part of understanding and applying that algorithm. For stochastic, adaptive heuristics in non-convex optimization, however, complexity analysis is often difficult. This article derives, for the first time in the literature, the complexity of the semi-steady-state Jaya algorithm (which is a recently developed variant of the Jaya algorithm) without the unimodality assumption. The Jaya algorithm, and its improvement, the semi-steady-state Jaya, are among the newest metaheuristics in population-based, nature-inspired optimization methods. In black-box function optimization, stochastic models of evolutionary and non-evolutionary heuristics often study the search process as sampling from distributions that are difficult to estimate. Unimodal distributions used for this purpose are easy to analyze but are necessarily restrictive. In this article, we model multimodality using mixtures of unimodal densities. For multimodal mixtures of uniform densities and, separately, of exponential densities (with different location parameters for the mixture components), analytical expressions, many of them closed-form, are derived for (i) the expectation of the largest order statistic for samples from the mixture; (ii) asymptotics of the above expectation for the large-sample case; (iii) survival probability corresponding to the (asymptotic) expected value of the largest order statistic; and (iv) asymptotics of sums of survival probabilities. The above quantities are used in a stochastic model of the semi-steady-state Jaya algorithm, obtaining the (asymptotic) expectation of the number of updates of the best individual in a population of the algorithm, which in turn is used in the derivation of the computational complexity of the algorithm.","PeriodicalId":13205,"journal":{"name":"IEEE Open Journal of the Computer Society","volume":"6 ","pages":"1390-1401"},"PeriodicalIF":0.0,"publicationDate":"2025-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11127039","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141629","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：