Journal of Information Security and Applications最新文献_第2页

Enhancing image steganography via frequency-guided iterative optimization 通过频率引导迭代优化增强图像隐写

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.jisa.2026.104391

Xinchen Wang , Duzhong Zhang , Jingwen Meng , Li Li , Taiyong Li

High-accuracy image steganography aims to conceal secret binary messages within a single cover image and recover them with minimal error. However, achieving this goal entails a fundamental trade-off: methods that excel in recovery often compromise visual quality and security. Existing one-shot deep learning approaches lack flexibility for fine-grained adjustment, whereas current iterative frameworks operate without perceptual guidance. Thus, both categories are limited in their ability to achieve accurate and imperceptible data embedding. To overcome these limitations, we propose a Frequency-Guided Iterative Network (FIS) that decouples embedding into two synergistic stages: iterative spatial refinement and explicit frequency-domain optimization. FIS comprises a flexible iterative encoder, a frequency perturbation module, and a decoder with a controlled obfuscation mechanism. The encoder iteratively refines the cover image to identify more suitable embedding locations, while the frequency perturbation module guides updates toward high-frequency regions where alterations are less perceptible. The decoder incorporates an obfuscation mechanism to enhance protection against unauthorized extraction. Experimental results across three datasets demonstrate that FIS achieves improved recovery accuracy, higher invisibility, and enhanced security.

高精度图像隐写术旨在将秘密二进制信息隐藏在单个封面图像中，并以最小的错误恢复它们。然而，实现这一目标需要一个基本的权衡：擅长恢复的方法通常会损害视觉质量和安全性。现有的一次性深度学习方法缺乏细粒度调整的灵活性，而当前的迭代框架在没有感知指导的情况下运行。因此，这两种类别在实现准确和难以察觉的数据嵌入方面的能力有限。为了克服这些限制，我们提出了一种频率引导迭代网络（FIS），它将嵌入解耦到两个协同阶段：迭代空间细化和显式频域优化。FIS包括一个灵活的迭代编码器、一个频率扰动模块和一个具有受控混淆机制的解码器。编码器迭代地细化封面图像以识别更合适的嵌入位置，而频率扰动模块将更新引导到变化不易察觉的高频区域。该解码器包含混淆机制以增强对未经授权的提取的保护。在三个数据集上的实验结果表明，FIS实现了更高的恢复精度、更高的不可见性和增强的安全性。

{"title":"Enhancing image steganography via frequency-guided iterative optimization","authors":"Xinchen Wang , Duzhong Zhang , Jingwen Meng , Li Li , Taiyong Li","doi":"10.1016/j.jisa.2026.104391","DOIUrl":"10.1016/j.jisa.2026.104391","url":null,"abstract":"<div><div>High-accuracy image steganography aims to conceal secret binary messages within a single cover image and recover them with minimal error. However, achieving this goal entails a fundamental trade-off: methods that excel in recovery often compromise visual quality and security. Existing one-shot deep learning approaches lack flexibility for fine-grained adjustment, whereas current iterative frameworks operate without perceptual guidance. Thus, both categories are limited in their ability to achieve accurate and imperceptible data embedding. To overcome these limitations, we propose a Frequency-Guided Iterative Network (FIS) that decouples embedding into two synergistic stages: iterative spatial refinement and explicit frequency-domain optimization. FIS comprises a flexible iterative encoder, a frequency perturbation module, and a decoder with a controlled obfuscation mechanism. The encoder iteratively refines the cover image to identify more suitable embedding locations, while the frequency perturbation module guides updates toward high-frequency regions where alterations are less perceptible. The decoder incorporates an obfuscation mechanism to enhance protection against unauthorized extraction. Experimental results across three datasets demonstrate that FIS achieves improved recovery accuracy, higher invisibility, and enhanced security.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104391"},"PeriodicalIF":3.7,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190373","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}

引用次数: 0

A TEE-based approach for preserving data secrecy in process mining with decentralized sources 一种基于tee的分散源过程挖掘数据保密性方法

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-02-02 DOI: 10.1016/j.jisa.2026.104381

Davide Basile , Valerio Goretti , Luca Barbaro , Hajo A. Reijers , Claudio Di Ciccio

Process mining techniques enable organizations to gain insights into their business processes through the analysis of execution records (event logs) stored by information systems. While most process mining efforts focus on intra-organizational scenarios, many real-world business processes span multiple independent organizations. Inter-organizational process mining, though, faces significant challenges, particularly regarding confidentiality guarantees: The analysis of data can reveal information that the participating organizations may not consent to disclose to one another, or to a third party hosting process mining services. To overcome this issue, this paper presents CONFINE, an approach for secrecy-preserving inter-organizational process mining. CONFINE leverages Trusted Execution Environments (TEEs) to deploy trusted applications that are capable of securely mining multi-party event logs while preserving data secrecy. We propose an architecture supporting a four-stage protocol to secure data exchange and processing, allowing for protected transfer and aggregation of unaltered process data across organizational boundaries. To avoid out-of-memory errors due to the limited capacity of TEEs, our protocol employs a segmentation-based strategy, whereby event logs are transmitted to TEEs in smaller batches. We conduct a formal verification of our approach’s correctness alongside a security analysis on the guarantees provided by the TEE core. We test our implementation using real-world and synthetic data to assess memory usage. Our experiments show that an incremental approach to segment processing in discovery and conformance checking is preferable over non-incremental strategies as the former maintains memory usage trends within a narrow range at runtime, whereas the latter exhibit high peaks towards the end of the execution. Furthermore, our results confirm that our prototype can handle real-world workloads without out-of-memory failures. The scalability tests reveal that memory usage grows logarithmically as the event log size increases. Memory consumption grows linearly with the number of provisioning organizations, indicating potential scalability limitations and opportunities for further optimizations.

流程挖掘技术使组织能够通过分析信息系统存储的执行记录（事件日志）来深入了解其业务流程。虽然大多数流程挖掘工作关注于组织内部场景，但许多现实世界的业务流程跨越多个独立的组织。但是，组织间流程挖掘面临着重大挑战，特别是在保密性保证方面：数据分析可能会揭示参与组织可能不同意向彼此或托管流程挖掘服务的第三方披露的信息。为了克服这个问题，本文提出了一种保密的组织间过程挖掘方法。restrict利用可信执行环境（tee）来部署可信的应用程序，这些应用程序能够在保护数据保密性的同时安全地挖掘多方事件日志。我们提出了一种支持四阶段协议的体系结构，以保护数据交换和处理，允许跨组织边界的未更改过程数据的受保护传输和聚合。为了避免由于tee容量有限而导致的内存不足错误，我们的协议采用了基于分段的策略，即以较小的批量将事件日志传输到tee。我们对方法的正确性进行正式验证，同时对TEE核心提供的保证进行安全性分析。我们使用真实世界和合成数据来测试我们的实现，以评估内存使用情况。我们的实验表明，在发现和一致性检查中，分段处理的增量方法比非增量策略更可取，因为前者在运行时将内存使用趋势保持在一个狭窄的范围内，而后者在执行结束时表现出高峰。此外，我们的结果证实，我们的原型可以处理实际工作负载，而不会出现内存不足故障。可伸缩性测试显示，随着事件日志大小的增加，内存使用量呈对数增长。内存消耗随着供应组织的数量呈线性增长，这表明了潜在的可伸缩性限制和进一步优化的机会。

{"title":"A TEE-based approach for preserving data secrecy in process mining with decentralized sources","authors":"Davide Basile , Valerio Goretti , Luca Barbaro , Hajo A. Reijers , Claudio Di Ciccio","doi":"10.1016/j.jisa.2026.104381","DOIUrl":"10.1016/j.jisa.2026.104381","url":null,"abstract":"<div><div>Process mining techniques enable organizations to gain insights into their business processes through the analysis of execution records (event logs) stored by information systems. While most process mining efforts focus on intra-organizational scenarios, many real-world business processes span multiple independent organizations. Inter-organizational process mining, though, faces significant challenges, particularly regarding confidentiality guarantees: The analysis of data can reveal information that the participating organizations may not consent to disclose to one another, or to a third party hosting process mining services. To overcome this issue, this paper presents CONFINE, an approach for secrecy-preserving inter-organizational process mining. CONFINE leverages Trusted Execution Environments (TEEs) to deploy trusted applications that are capable of securely mining multi-party event logs while preserving data secrecy. We propose an architecture supporting a four-stage protocol to secure data exchange and processing, allowing for protected transfer and aggregation of unaltered process data across organizational boundaries. To avoid out-of-memory errors due to the limited capacity of TEEs, our protocol employs a segmentation-based strategy, whereby event logs are transmitted to TEEs in smaller batches. We conduct a formal verification of our approach’s correctness alongside a security analysis on the guarantees provided by the TEE core. We test our implementation using real-world and synthetic data to assess memory usage. Our experiments show that an incremental approach to segment processing in discovery and conformance checking is preferable over non-incremental strategies as the former maintains memory usage trends within a narrow range at runtime, whereas the latter exhibit high peaks towards the end of the execution. Furthermore, our results confirm that our prototype can handle real-world workloads without out-of-memory failures. The scalability tests reveal that memory usage grows logarithmically as the event log size increases. Memory consumption grows linearly with the number of provisioning organizations, indicating potential scalability limitations and opportunities for further optimizations.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104381"},"PeriodicalIF":3.7,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190388","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}

引用次数: 0

MFTA-PFL : Multi-factor trust assessment-based personalized federated learning 基于多因素信任评估的个性化联邦学习

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-02-01 DOI: 10.1016/j.jisa.2026.104388

Fahad Sabah , Yuwen Chen , Zhen Yang , Muhammad Azam , Nadeem Ahmad , Raheem Sarwar

Federated Learning (FL) has emerged as a promising approach for decentralized machine learning, enabling multiple clients to collaboratively train a global model without sharing their local data. However, challenges such as data heterogeneity, varying client reliability, and communication constraints often hinder model performance and convergence. While existing methods address aspects like resource efficiency or reputation-based selection, few integrate multi-dimensional trust evaluation in a unified framework. To bridge this gap, this paper proposes a Multi-Factor Trust Assessment based Personalized Federated Learning (MFTA-PFL). Our approach dynamically evaluates clients based on model accuracy, data quality, historical reliability, and communication metrics, assigning trust scores that determine their selection priority. By preferentially engaging higher-trust clients, MFTA-PFL enhances the robustness and efficiency of the FL process. Extensive experiments on non-IID versions of MNIST and Fashion-MNIST demonstrate that our method outperforms conventional client selection strategies, achieving superior accuracy; 98.84% on MNIST and 88.82% on Fashion-MNIST, faster convergence, and improved communication efficiency-even under adversarial conditions. These results highlight the critical role of adaptive, trust-aware client selection in building scalable and reliable FL systems.

联邦学习（FL）已经成为分散机器学习的一种有前途的方法，使多个客户端能够在不共享本地数据的情况下协作训练全局模型。然而，诸如数据异构性、不同的客户端可靠性和通信约束等挑战通常会阻碍模型的性能和收敛。虽然现有的方法解决资源效率或基于声誉的选择等方面，但很少将多维信任评估整合到统一的框架中。为此，本文提出了一种基于多因素信任评估的个性化联邦学习（MFTA-PFL）方法。我们的方法基于模型准确性、数据质量、历史可靠性和通信指标动态评估客户，分配信任分数，确定客户的选择优先级。通过优先与高信任度的客户合作，MFTA-PFL提高了FL流程的稳健性和效率。在非iid版本的MNIST和Fashion-MNIST上进行的大量实验表明，我们的方法优于传统的客户选择策略，实现了更高的准确性；MNIST为98.84%，Fashion-MNIST为88.82%，更快的融合，甚至在敌对条件下也提高了通信效率。这些结果突出了自适应、信任意识客户选择在构建可扩展和可靠的FL系统中的关键作用。

{"title":"MFTA-PFL : Multi-factor trust assessment-based personalized federated learning","authors":"Fahad Sabah , Yuwen Chen , Zhen Yang , Muhammad Azam , Nadeem Ahmad , Raheem Sarwar","doi":"10.1016/j.jisa.2026.104388","DOIUrl":"10.1016/j.jisa.2026.104388","url":null,"abstract":"<div><div>Federated Learning (FL) has emerged as a promising approach for decentralized machine learning, enabling multiple clients to collaboratively train a global model without sharing their local data. However, challenges such as data heterogeneity, varying client reliability, and communication constraints often hinder model performance and convergence. While existing methods address aspects like resource efficiency or reputation-based selection, few integrate multi-dimensional trust evaluation in a unified framework. To bridge this gap, this paper proposes a Multi-Factor Trust Assessment based Personalized Federated Learning (MFTA-PFL). Our approach dynamically evaluates clients based on model accuracy, data quality, historical reliability, and communication metrics, assigning trust scores that determine their selection priority. By preferentially engaging higher-trust clients, MFTA-PFL enhances the robustness and efficiency of the FL process. Extensive experiments on non-IID versions of MNIST and Fashion-MNIST demonstrate that our method outperforms conventional client selection strategies, achieving superior accuracy; 98.84% on MNIST and 88.82% on Fashion-MNIST, faster convergence, and improved communication efficiency-even under adversarial conditions. These results highlight the critical role of adaptive, trust-aware client selection in building scalable and reliable FL systems.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104388"},"PeriodicalIF":3.7,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190385","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}

引用次数: 0

Zerovision: A privacy-preserving iris authentication framework using zero knowledge proofs and steganographic safeguards Zerovision：一个保护隐私的虹膜认证框架，使用零知识证明和隐写保护

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-29 DOI: 10.1016/j.jisa.2025.104323

Khushil Godhani , Nihhar Shukla , Janam Patel , Rajesh Gupta , Sudeep Tanwar

Biometric authentication systems, particularly those relying on iris recognition, offer an extremely accurate and secure method of identity verification, but the very fact that such an industry exists has raised issues regarding individual privacy. Biometric data stolen from a system, unlike passwords, cannot be replaced and can be used for identity theft. This paper presents ZeroVision, a novel privacy-preserving iris authentication scheme with a blend of steganography, convolutional neural networks (CNNs), zero-knowledge proofs (zk-SNARKs), and blockchain. ZeroVision conceals iris images in cover facial images through steganography to hide their transmission and provoke transmission security. CNNs are utilized to obtain compact binary feature templates from iris image, whereas zk-SNARKs allow verifiers to authenticate template validity in zero knowledge, which keeps any sensitive information disclosure distant. Blockchain deployment guarantees that the proofs generated are accurate, verified by the verifier, and stored in a decentralized, tamper-proof fashion. Tested on the CASIA Iris Thousand and FFHQ datasets in a simulation of real-world transactions and transmissions, ZeroVision attains 91.41 % accuracy for recognition despite compact template sizes and additional noise, with proof generation and verification times of under 0.6 and 0.25 seconds, respectively. This novel architecture enables secure biometric authentication in high-risk applications where the privacy of personal data is highest priority.

生物识别认证系统，特别是那些依赖虹膜识别的系统，提供了一种极其准确和安全的身份验证方法，但事实上，这样一个行业的存在引发了有关个人隐私的问题。与密码不同，从系统中窃取的生物识别数据无法替换，可用于身份盗窃。ZeroVision是一种新型的保护隐私的虹膜认证方案，它融合了隐写术、卷积神经网络（cnn）、零知识证明（zk- snark）和区块链。ZeroVision通过隐写术将虹膜图像隐藏在人脸图像中，以隐藏其传输，提高传输安全性。利用cnn从虹膜图像中获得紧凑的二值特征模板，而zk-SNARKs允许验证者在零知识的情况下验证模板的有效性，从而避免任何敏感信息泄露。区块链部署保证生成的证明是准确的，由验证者验证，并以分散的、防篡改的方式存储。在模拟真实世界交易和传输的CASIA Iris Thousand和FFHQ数据集上进行测试，尽管模板尺寸紧凑且存在额外的噪声，但ZeroVision的识别准确率达到了91.41%，证明生成和验证时间分别低于0.6秒和0.25秒。这种新颖的体系结构可以在个人数据隐私最高优先级的高风险应用中实现安全的生物识别认证。

{"title":"Zerovision: A privacy-preserving iris authentication framework using zero knowledge proofs and steganographic safeguards","authors":"Khushil Godhani , Nihhar Shukla , Janam Patel , Rajesh Gupta , Sudeep Tanwar","doi":"10.1016/j.jisa.2025.104323","DOIUrl":"10.1016/j.jisa.2025.104323","url":null,"abstract":"<div><div>Biometric authentication systems, particularly those relying on iris recognition, offer an extremely accurate and secure method of identity verification, but the very fact that such an industry exists has raised issues regarding individual privacy. Biometric data stolen from a system, unlike passwords, cannot be replaced and can be used for identity theft. This paper presents <em>ZeroVision</em>, a novel privacy-preserving iris authentication scheme with a blend of steganography, convolutional neural networks (CNNs), zero-knowledge proofs (zk-SNARKs), and blockchain. <em>ZeroVision</em> conceals iris images in cover facial images through steganography to hide their transmission and provoke transmission security. CNNs are utilized to obtain compact binary feature templates from iris image, whereas zk-SNARKs allow verifiers to authenticate template validity in zero knowledge, which keeps any sensitive information disclosure distant. Blockchain deployment guarantees that the proofs generated are accurate, verified by the verifier, and stored in a decentralized, tamper-proof fashion. Tested on the CASIA Iris Thousand and FFHQ datasets in a simulation of real-world transactions and transmissions, <em>ZeroVision</em> attains 91.41 % accuracy for recognition despite compact template sizes and additional noise, with proof generation and verification times of under 0.6 and 0.25 seconds, respectively. This novel architecture enables secure biometric authentication in high-risk applications where the privacy of personal data is highest priority.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104323"},"PeriodicalIF":3.7,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080983","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}

引用次数: 0

AMF-CFL: Anomaly model filtering based on clustering in federated learning AMF-CFL：联邦学习中基于聚类的异常模型过滤

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-29 DOI: 10.1016/j.jisa.2026.104387

Bo Wang , Xiaorui Dai , Wei Wang , Zi Yang , Zhaoning Wang , Maozhen Zhang

Federated learning (FL) allows multiple participants to collaboratively train a shared model without exposing their local data, thereby mitigating the risk of data leakage. Despite its advantages, FL is vulnerable to attacks by malicious clients, and existing defense mechanisms, while effective under independent and identically distributed (i.i.d.) settings, often exhibit degraded performance in non-i.i.d. scenarios where client data distributions differ. To overcome this limitation, we propose AMF-CFL, a robust aggregation algorithm specifically designed for federated learning under non-i.i.d. conditions. AMF-CFL reduces the influence of malicious updates through a two-step filtering strategy: it first applies multi-k-means clustering to identify anomalous update patterns, followed by z-score-based statistical analysis to refine the selection of benign updates. Comprehensive evaluations against four untargeted and two targeted attack types demonstrate that AMF-CFL effectively preserves the integrity and robustness of the global model, offering a reliable defense in challenging federated learning environments.

联邦学习（FL）允许多个参与者在不暴露本地数据的情况下协作训练共享模型，从而降低了数据泄漏的风险。尽管具有优势，但FL很容易受到恶意客户端的攻击，现有的防御机制虽然在独立和同分布（i.i.d）设置下有效，但在非i.i.d设置下往往表现出性能下降。客户端数据分布不同的场景。为了克服这一限制，我们提出了AMF-CFL算法，这是一种专门为非id下的联邦学习设计的鲁棒聚合算法。条件。AMF-CFL通过两步过滤策略来减少恶意更新的影响：首先应用多k均值聚类来识别异常更新模式，然后使用基于z分数的统计分析来优化良性更新的选择。针对四种非目标攻击和两种目标攻击类型的综合评估表明，AMF-CFL有效地保持了全局模型的完整性和鲁棒性，在具有挑战性的联邦学习环境中提供了可靠的防御。

{"title":"AMF-CFL: Anomaly model filtering based on clustering in federated learning","authors":"Bo Wang , Xiaorui Dai , Wei Wang , Zi Yang , Zhaoning Wang , Maozhen Zhang","doi":"10.1016/j.jisa.2026.104387","DOIUrl":"10.1016/j.jisa.2026.104387","url":null,"abstract":"<div><div>Federated learning (FL) allows multiple participants to collaboratively train a shared model without exposing their local data, thereby mitigating the risk of data leakage. Despite its advantages, FL is vulnerable to attacks by malicious clients, and existing defense mechanisms, while effective under independent and identically distributed (i.i.d.) settings, often exhibit degraded performance in non-i.i.d. scenarios where client data distributions differ. To overcome this limitation, we propose AMF-CFL, a robust aggregation algorithm specifically designed for federated learning under non-i.i.d. conditions. AMF-CFL reduces the influence of malicious updates through a two-step filtering strategy: it first applies multi-<em>k</em>-means clustering to identify anomalous update patterns, followed by <em>z</em>-score-based statistical analysis to refine the selection of benign updates. Comprehensive evaluations against four untargeted and two targeted attack types demonstrate that AMF-CFL effectively preserves the integrity and robustness of the global model, offering a reliable defense in challenging federated learning environments.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104387"},"PeriodicalIF":3.7,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080917","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}

引用次数: 0

Say the image: Auditory masking effect-driven invertible network for progressive image-in-audio steganography 说图像：听觉掩蔽效应驱动的渐进图像音频隐写的可逆网络

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-27 DOI: 10.1016/j.jisa.2026.104382

Jinghang Song , Fangyuan Gao , Xin Deng , Shengxi Li , Mai Xu

In this paper, we propose an auditory masking effect-driven invertible network for Hiding an Image within an Audio signal, termed as HIA-Net. Unlike the direct hiding manner, the proposed HIA-Net decomposes the image-in-audio steganography process into two cascaded stages. In the first stage, we develop a Masker Audio Extraction (MAE) algorithm to turn the original cover audio into a masker audio. The generated masker audio exhibits higher masking capability, thereby enhancing the hiding invisibility and security. Then, we design three Image-in-Audio Invertible (I-AI) sub-networks to embed the secret image into the masker audio, yielding a stego masker audio. In the second stage, an Audio-in-Audio Invertible (A-AI) sub-network is employed to further conceal the stego masker audio within the original cover audio, producing the final stego audio. During the revealing process, the reversible architecture of the proposed network first reconstructs the stego masker from the final stego audio, and then recovers the hidden image from the stego masker. Experimental results demonstrate that HIA-Net significantly outperforms other state-of-the-art image-in-audio steganography methods, achieving a significant PSNR improvement of more than 3.0 dB for secret image reconstruction on different image and audio datasets. The user study also confirms the superior imperceptibility of the stego audios. The software code is available at https://github.com/c4Tch3r/HIANet.

在本文中，我们提出了一个听觉掩蔽效应驱动的可逆网络，用于隐藏音频信号中的图像，称为HIA-Net。与直接隐藏方式不同，本文提出的HIA-Net将图像-音频隐写过程分解为两个级联阶段。在第一阶段，我们开发了一种掩蔽音频提取（MAE）算法，将原始掩蔽音频转换为掩蔽音频。所生成的掩码音频具有更高的掩码能力，从而增强了隐藏的不可见性和安全性。然后，我们设计了三个图像-音频可逆（I-AI）子网络，将秘密图像嵌入到掩蔽器音频中，产生一个隐去掩蔽器音频。在第二阶段，使用音频中音频可逆（A-AI）子网络进一步隐藏原始掩蔽音频中的隐写掩蔽音频，产生最终的隐写音频。在揭示过程中，该网络的可逆结构首先从最终的隐去音频中重建隐去掩模，然后从隐去掩模中恢复隐藏图像。实验结果表明，HIA-Net显著优于其他最先进的图像音频隐写方法，在不同的图像和音频数据集上实现了超过3.0 dB的秘密图像重建的显着PSNR提高。用户研究也证实了隐音音频优越的隐蔽性。软件代码可在https://github.com/c4Tch3r/HIANet上获得。

{"title":"Say the image: Auditory masking effect-driven invertible network for progressive image-in-audio steganography","authors":"Jinghang Song , Fangyuan Gao , Xin Deng , Shengxi Li , Mai Xu","doi":"10.1016/j.jisa.2026.104382","DOIUrl":"10.1016/j.jisa.2026.104382","url":null,"abstract":"<div><div>In this paper, we propose an auditory masking effect-driven invertible network for Hiding an Image within an Audio signal, termed as HIA-Net. Unlike the direct hiding manner, the proposed HIA-Net decomposes the image-in-audio steganography process into two cascaded stages. In the first stage, we develop a Masker Audio Extraction (MAE) algorithm to turn the original cover audio into a masker audio. The generated masker audio exhibits higher masking capability, thereby enhancing the hiding invisibility and security. Then, we design three Image-in-Audio Invertible (I-AI) sub-networks to embed the secret image into the masker audio, yielding a stego masker audio. In the second stage, an Audio-in-Audio Invertible (A-AI) sub-network is employed to further conceal the stego masker audio within the original cover audio, producing the final stego audio. During the revealing process, the reversible architecture of the proposed network first reconstructs the stego masker from the final stego audio, and then recovers the hidden image from the stego masker. Experimental results demonstrate that HIA-Net significantly outperforms other state-of-the-art image-in-audio steganography methods, achieving a significant PSNR improvement of more than 3.0 dB for secret image reconstruction on different image and audio datasets. The user study also confirms the superior imperceptibility of the stego audios. The software code is available at <span><span>https://github.com/c4Tch3r/HIANet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104382"},"PeriodicalIF":3.7,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080981","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}

引用次数: 0

FedCPP: A hybrid proactive-passive defense framework for backdoor attack mitigation in federated learning FedCPP：用于联邦学习中后门攻击缓解的混合主动-被动防御框架

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-27 DOI: 10.1016/j.jisa.2026.104377

Longhang Xu , Guanxin Chen , Nan Yang , Yipen Liu , Jianting Yuan

Federated learning (FL), as a collaborative model training paradigm among multiple clients, is inherently susceptible to backdoor attacks due to its privacy-preserving requirements. In such attacks, adversaries embed triggers into the global model, causing it to produce targeted output when encountering specific inputs. Existing defense mechanisms are generally categorized into proactive and passive strategies. Proactive strategies, such as differential privacy and noise injection, can slightly alleviate the impact of backdoor attacks but often degrade model performance. Passive strategies, which rely on distance or similarity to detect, typically assume ideal conditions and impose strict constraints on the attacker’s data distribution and the number of malicious clients. To address these limitations, we propose FedCPP, an effective defense framework that combines the strengths of both proactive and passive strategies. Specifically, FedCPP first employs a proactive mechanism to identify critical layers targeted by backdoor attacks. It then integrates a passive defense strategy based on multi-metric evaluation, coupled with a dynamic weighted adaptive algorithm to achieve defense against backdoor attacks. Experimental results demonstrate that FedCPP effectively detects backdoor attacks in FL scenarios without constraints on the proportion of malicious participants, data distribution, or attack timing while maintaining high model accuracy. Compared to existing state-of-the-art defensive strategies, FedCPP achieves superior performance with minimal impact on the global model.

联邦学习（FL）作为多客户机之间的协作模型训练范例，由于其隐私保护要求，本质上容易受到后门攻击。在这种攻击中，攻击者将触发器嵌入到全局模型中，使其在遇到特定输入时产生目标输出。现有的防御机制一般分为主动防御和被动防御。主动策略，如差分隐私和噪声注入，可以稍微减轻后门攻击的影响，但通常会降低模型性能。被动策略依赖距离或相似性进行检测，通常假设理想条件，并对攻击者的数据分布和恶意客户端数量施加严格约束。为了解决这些限制，我们提出了FedCPP，这是一个有效的防御框架，结合了主动和被动战略的优势。具体来说，FedCPP首先采用了一种主动机制来识别后门攻击的关键层。然后，将基于多度量评估的被动防御策略与动态加权自适应算法相结合，实现对后门攻击的防御。实验结果表明，FedCPP可以有效检测FL场景下的后门攻击，不受恶意参与者比例、数据分布、攻击时间的限制，同时保持较高的模型精度。与现有的最先进的防御策略相比，FedCPP在对全局模型影响最小的情况下实现了卓越的性能。

{"title":"FedCPP: A hybrid proactive-passive defense framework for backdoor attack mitigation in federated learning","authors":"Longhang Xu , Guanxin Chen , Nan Yang , Yipen Liu , Jianting Yuan","doi":"10.1016/j.jisa.2026.104377","DOIUrl":"10.1016/j.jisa.2026.104377","url":null,"abstract":"<div><div>Federated learning (FL), as a collaborative model training paradigm among multiple clients, is inherently susceptible to backdoor attacks due to its privacy-preserving requirements. In such attacks, adversaries embed triggers into the global model, causing it to produce targeted output when encountering specific inputs. Existing defense mechanisms are generally categorized into proactive and passive strategies. Proactive strategies, such as differential privacy and noise injection, can slightly alleviate the impact of backdoor attacks but often degrade model performance. Passive strategies, which rely on distance or similarity to detect, typically assume ideal conditions and impose strict constraints on the attacker’s data distribution and the number of malicious clients. To address these limitations, we propose FedCPP, an effective defense framework that combines the strengths of both proactive and passive strategies. Specifically, FedCPP first employs a proactive mechanism to identify critical layers targeted by backdoor attacks. It then integrates a passive defense strategy based on multi-metric evaluation, coupled with a dynamic weighted adaptive algorithm to achieve defense against backdoor attacks. Experimental results demonstrate that FedCPP effectively detects backdoor attacks in FL scenarios without constraints on the proportion of malicious participants, data distribution, or attack timing while maintaining high model accuracy. Compared to existing state-of-the-art defensive strategies, FedCPP achieves superior performance with minimal impact on the global model.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104377"},"PeriodicalIF":3.7,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080982","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}

引用次数: 0

Decoupled framework for non-additive adversarial image steganography 非加性对抗图像隐写的解耦框架

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-27 DOI: 10.1016/j.jisa.2026.104376

Junfeng Zhao , Shen Wang

Adversarial image steganography aims to introduce a small amount of perturbations during the data embedding to improve security performance, while existing works are typically based on additive model under the framework of distortion minimization. Different from additive model, non-additive model assumes that the modification of adjacent elements will interact with each other. If adversarial perturbations are introduced on this basis, the performance of adversarial stegos against re-trained steganalyzers will be further improved. In this paper, we point out the reasons why the existing coupled framework causes the actual embedding structure to fail to fully meet the constraints of the non-additive embedding structure. Then, we decouple the two methods according to their roles, making them independent in structure and more flexible in combination. However, since non-additive adversarial image steganography have to follow the constraints, if the steganographer still aims to successfully attack the target model, excessive perturbations will be occurred. To avoid this phenomenon, we propose a mechanism based on the difference in the attack threshold between the two methods. Extensive experimental results show that if the steganographer uses the decoupled framework to reconstruct the methods, an adversarial stego that satisfies the non-additive constraints can be generated, and the security performance against re-trained steganalyzers in the spatial domain is improved by about 1% ~3% compared with the additive model-based method.

对抗图像隐写的目的是在数据嵌入过程中引入少量的扰动以提高安全性能，而现有的工作通常是基于失真最小化框架下的加性模型。与加性模型不同，非加性模型假设相邻元素的修改会相互作用。如果在此基础上引入对抗性扰动，对抗性隐写算法对重新训练的隐写分析器的性能将进一步提高。本文指出了现有耦合框架导致实际嵌入结构不能完全满足非加性嵌入结构约束的原因。然后，我们根据两种方法的作用进行解耦，使它们在结构上独立，在组合上更加灵活。然而，由于非加性对抗性图像隐写必须遵循约束，如果隐写者仍然以成功攻击目标模型为目标，则会产生过多的扰动。为了避免这种现象，我们提出了一种基于两种方法攻击阈值差异的机制。大量的实验结果表明，如果隐写者使用解耦框架重构方法，可以生成满足非加性约束的对抗隐写，并且与基于加性模型的方法相比，在空间域中对重新训练的隐写分析器的安全性能提高了约1% ~3%。

{"title":"Decoupled framework for non-additive adversarial image steganography","authors":"Junfeng Zhao , Shen Wang","doi":"10.1016/j.jisa.2026.104376","DOIUrl":"10.1016/j.jisa.2026.104376","url":null,"abstract":"<div><div>Adversarial image steganography aims to introduce a small amount of perturbations during the data embedding to improve security performance, while existing works are typically based on additive model under the framework of distortion minimization. Different from additive model, non-additive model assumes that the modification of adjacent elements will interact with each other. If adversarial perturbations are introduced on this basis, the performance of adversarial stegos against re-trained steganalyzers will be further improved. In this paper, we point out the reasons why the existing coupled framework causes the actual embedding structure to fail to fully meet the constraints of the non-additive embedding structure. Then, we decouple the two methods according to their roles, making them independent in structure and more flexible in combination. However, since non-additive adversarial image steganography have to follow the constraints, if the steganographer still aims to successfully attack the target model, excessive perturbations will be occurred. To avoid this phenomenon, we propose a mechanism based on the difference in the attack threshold between the two methods. Extensive experimental results show that if the steganographer uses the decoupled framework to reconstruct the methods, an adversarial stego that satisfies the non-additive constraints can be generated, and the security performance against re-trained steganalyzers in the spatial domain is improved by about 1% ~3% compared with the additive model-based method.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104376"},"PeriodicalIF":3.7,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080984","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}

引用次数: 0

A secured cryptographic approach with extreme gradient boosting model for data aggregation and routing in WSN 基于极端梯度增强模型的无线传感器网络数据聚合和路由安全加密方法

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-24 DOI: 10.1016/j.jisa.2026.104372

Ramkumar Devendiran, Anil V Turukmane

An effective routing algorithm is responsible for the efficiency and security of wireless sensor networks (WSNs). There have been numerous investigations during recent periods that have attempted to enhance route security, data confidentiality, and data aggregation processes. These techniques have some significant disadvantages, including data loss, expensive temporal complexity, and vulnerability to different kinds of attacks (e.g., passive, malevolent, or aggressive attacks). The objective of this study is to develop a machine-learning algorithm for secure data aggregation and an encryption algorithm for secure routing in WSNs. Sensor nodes are initially deployed in a WSN, and nodes are thereafter grouped according to the Modified Fuzzy C-Means Clustering (MFCMC) algorithm. Subsequently, the node aggregates the data using the Extreme Gradient Boosting (XGBoost) algorithm. Thereafter, encryption is carried out using the application of the Feistel Shaped Tiny Encryption (FSTE) technique. Lastly, encrypted data is passed through a novel Opposition Learning based Honey Badger Optimization (OL_HBO) technique to choose the best route. This approach is based on parameters such as residual energy, node degree, node centrality, and distance between sensor nodes. In an evaluation setting, the proposed technique achieves an average end-to-end delay (58.73 ms), packet delivery ratio (PDR) (90.37%), throughput (253.41 kbps), encryption time (0.39 ms), and decryption time (6.1 ms). By comparing the performance of the proposed technique with other state-of-the-art approaches, the results demonstrate improved performance.

有效的路由算法对无线传感器网络的效率和安全性起着至关重要的作用。在最近的一段时间里，有许多研究试图增强路由安全性、数据机密性和数据聚合过程。这些技术有一些明显的缺点，包括数据丢失、昂贵的时间复杂性以及容易受到不同类型的攻击（例如被动攻击、恶意攻击或攻击性攻击）。本研究的目的是开发一种用于安全数据聚合的机器学习算法和一种用于WSNs安全路由的加密算法。传感器节点最初部署在WSN中，然后根据改进模糊c均值聚类（MFCMC）算法对节点进行分组。随后，节点使用极限梯度增强（XGBoost）算法聚合数据。然后，利用费斯特尔形状微小加密（FSTE）技术进行加密。最后，通过一种新颖的基于对立学习的蜂蜜獾优化（OL_HBO）技术来传递加密数据，以选择最佳路径。该方法基于剩余能量、节点度、节点中心性和传感器节点之间的距离等参数。在评估设置中，所提出的技术实现了平均端到端延迟（58.73 ms）、分组传输比（PDR）（90.37%）、吞吐量（253.41 kbps）、加密时间（0.39 ms）和解密时间（6.1 ms）。通过将所提出的技术性能与其他最先进的方法进行比较，结果表明性能有所提高。

{"title":"A secured cryptographic approach with extreme gradient boosting model for data aggregation and routing in WSN","authors":"Ramkumar Devendiran, Anil V Turukmane","doi":"10.1016/j.jisa.2026.104372","DOIUrl":"10.1016/j.jisa.2026.104372","url":null,"abstract":"<div><div>An effective routing algorithm is responsible for the efficiency and security of wireless sensor networks (WSNs). There have been numerous investigations during recent periods that have attempted to enhance route security, data confidentiality, and data aggregation processes. These techniques have some significant disadvantages, including data loss, expensive temporal complexity, and vulnerability to different kinds of attacks (e.g., passive, malevolent, or aggressive attacks). The objective of this study is to develop a machine-learning algorithm for secure data aggregation and an encryption algorithm for secure routing in WSNs. Sensor nodes are initially deployed in a WSN, and nodes are thereafter grouped according to the Modified Fuzzy C-Means Clustering (MFCMC) algorithm. Subsequently, the node aggregates the data using the Extreme Gradient Boosting (XGBoost) algorithm. Thereafter, encryption is carried out using the application of the Feistel Shaped Tiny Encryption (FSTE) technique. Lastly, encrypted data is passed through a novel Opposition Learning based Honey Badger Optimization (OL_HBO) technique to choose the best route. This approach is based on parameters such as residual energy, node degree, node centrality, and distance between sensor nodes. In an evaluation setting, the proposed technique achieves an average end-to-end delay (58.73 ms), packet delivery ratio (PDR) (90.37%), throughput (253.41 kbps), encryption time (0.39 ms), and decryption time (6.1 ms). By comparing the performance of the proposed technique with other state-of-the-art approaches, the results demonstrate improved performance.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104372"},"PeriodicalIF":3.7,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039809","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}

引用次数: 0

AGentVLM: Access control policy generation and verification framework with language models AGentVLM：带有语言模型的访问控制策略生成和验证框架

IF 3.7 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Journal of Information Security and Applications

Pub Date : 2026-01-23 DOI: 10.1016/j.jisa.2026.104379

Sakuna Harinda Jayasundara , Nalin Asanka Gamagedara Arachchilage , Giovanni Russello

Manual generation of access control policies from high-level organizational requirements is labor-intensive and error-prone, often leading to critical failures and data breaches. While automated frameworks have been proposed, existing approaches struggle with complex access requirements due to poor domain adaptation, limiting their accuracy. To address these challenges, we propose AGentVLM, a novel access control policy generation and verification framework based on small, open-source language models (LMs). Our framework enables its efficient on-premise deployment, preserving data confidentiality by avoiding reliance on third-party black-box LMs. AGentVLM excels in identifying natural language access control policies (NLACPs) from high-level requirements, achieving an average F1 score of 90.6 %. Unlike existing frameworks limited to generating simple policies with three components (subject, action, resource), AGentVLM effectively extracts complex elements such as purposes and conditions using an access control-specific structured information extraction technique. This method captures both word-level and semantic information at the same time from NLACPs, leading to a state-of-the-art policy generation F1 score of 80.6 %. Additionally, AGentVLM introduces a verification technique that provides actionable feedback, allowing administrators to refine inaccurate policies before deployment. To support future research, we also release two annotated datasets addressing the scarcity of domain-specific data.

从高级组织需求手动生成访问控制策略是一项劳动密集型工作，而且容易出错，经常导致严重故障和数据泄露。虽然已经提出了自动化框架，但现有的方法由于域适应性差而难以满足复杂的访问需求，从而限制了它们的准确性。为了解决这些挑战，我们提出了AGentVLM，这是一种基于小型开源语言模型（LMs）的新型访问控制策略生成和验证框架。我们的框架实现了高效的内部部署，避免了对第三方黑匣子lm的依赖，从而保护了数据的机密性。AGentVLM擅长从高级需求中识别自然语言访问控制策略（nlacp），平均F1得分为90.6%。与仅限于生成具有三个组件（主题、操作、资源）的简单策略的现有框架不同，AGentVLM使用特定于访问控制的结构化信息提取技术有效地提取复杂元素，如目的和条件。该方法同时从nlacp中捕获词级和语义信息，导致最先进的策略生成F1得分为80.6%。此外，AGentVLM引入了一种验证技术，该技术提供了可操作的反馈，允许管理员在部署之前改进不准确的策略。为了支持未来的研究，我们还发布了两个带注释的数据集，以解决特定领域数据的稀缺性。

{"title":"AGentVLM: Access control policy generation and verification framework with language models","authors":"Sakuna Harinda Jayasundara , Nalin Asanka Gamagedara Arachchilage , Giovanni Russello","doi":"10.1016/j.jisa.2026.104379","DOIUrl":"10.1016/j.jisa.2026.104379","url":null,"abstract":"<div><div>Manual generation of access control policies from high-level organizational requirements is labor-intensive and error-prone, often leading to critical failures and data breaches. While automated frameworks have been proposed, existing approaches struggle with complex access requirements due to poor domain adaptation, limiting their accuracy. To address these challenges, we propose AGentVLM, a novel access control policy generation and verification framework based on small, open-source language models (LMs). Our framework enables its efficient on-premise deployment, preserving data confidentiality by avoiding reliance on third-party black-box LMs. AGentVLM excels in identifying natural language access control policies (NLACPs) from high-level requirements, achieving an average F1 score of 90.6 %. Unlike existing frameworks limited to generating simple policies with three components (subject, action, resource), AGentVLM effectively extracts complex elements such as purposes and conditions using an access control-specific structured information extraction technique. This method captures both word-level and semantic information at the same time from NLACPs, leading to a state-of-the-art policy generation F1 score of 80.6 %. Additionally, AGentVLM introduces a verification technique that provides actionable feedback, allowing administrators to refine inaccurate policies before deployment. To support future research, we also release two annotated datasets addressing the scarcity of domain-specific data.</div></div>","PeriodicalId":48638,"journal":{"name":"Journal of Information Security and Applications","volume":"98 ","pages":"Article 104379"},"PeriodicalIF":3.7,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039808","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}

引用次数: 0