Junwei Zhou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Junwei Zhou has a total of 118 co-authors. Here are the top co-authors by number of shared publications:
- Jianwen Xiang
Wuhan University of Technology, Wuhan, China
10 shared publications Jianwen Xiang
5 shared publications- Kaitai Liang
Department of Computing, University of Turku, Turku, Finland
4 shared publications - Qiuzhen Lin
Shenzhen University
4 shared publications Yanchao Yang
Wuhan University of Technology
4 shared publications- Cheng Tan
Wuhan University of Technology
3 shared publications
Latest Publications
- Open Access
- 1 September 2026
- Published by Elsevier in SSRN Electronic Journal
CrossBot: Android GUI Testing via Efficient Cross-Device History Reuse
Context: Android applications are increasingly released as multiple device-specific variants, such as phone, tablet, in-vehicle, and TV versions. Existing automated GUI testing tools usually test these variants independently, repeatedly spending testing budget on common functionality and providing little guidance when a new phone variant is explored.Objectives: This paper aims to improve phone-target Android GUI testing by reusing testing histories collected offline from non-phone variants of the same app, while remaining robust to cross-device differences in layouts, widgets, and interaction flows.Methods: We present \texttt{CrossBot}, a widget-level history reuse framework. \texttt{CrossBot} summarizes source-side testing histories as probabilistic event-to-widget transition priors, matches semantically corresponding widgets across device variants using text, resource-id, and activity context, and injects the transferred priors into online phone-side exploration. To reduce online overhead, \texttt{CrossBot} further applies embedding caching and locality-sensitive hashing. We evaluate \texttt{CrossBot} on 14 commercial Android apps under Tablet-to-Phone, TV-to-Phone, Car-to-Phone, and Multi-to-Phone reuse settings.Results: Compared with \texttt{Fastbot2}, \texttt{CrossBot-With-X} improves activity coverage by 16.5\%--24.2\% and widget coverage by 62.0\%--88.5\% at 60 minutes. It also outperforms a resource-ID-only reuse baseline and sustains coverage growth in four-hour sessions. In the LSH ablation, LSH-backed matching increases throughput by 97.3\% and reduces amortized cost by 46.8\% on average.Conclusion: The results show that widget-level cross-device history reuse can effectively guide phone-target GUI exploration under UI differences, and that efficient approximate matching makes such reuse practical for long-running Android testing.
- 1 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCrossBot: Android GUI Testing via Efficient Cross-Device History Reuse
Context: Android applications are increasingly released as multiple device-specific variants, such as phone, tablet, in-vehicle, and TV versions. Existing automated GUI testing tools usually test these variants independently, repeatedly spending testing budget on common functionality and providing little guidance when a new phone variant is explored.Objectives: This paper aims to improve phone-target Android GUI testing by reusing testing histories collected offline from non-phone variants of the same app, while remaining robust to cross-device differences in layouts, widgets, and interaction flows.Methods: We present \texttt{CrossBot}, a widget-level history reuse framework. \texttt{CrossBot} summarizes source-side testing histories as probabilistic event-to-widget transition priors, matches semantically corresponding widgets across device variants using text, resource-id, and activity context, and injects the transferred priors into online phone-side exploration. To reduce online overhead, \texttt{CrossBot} further applies embedding caching and locality-sensitive hashing. We evaluate \texttt{CrossBot} on 14 commercial Android apps under Tablet-to-Phone, TV-to-Phone, Car-to-Phone, and Multi-to-Phone reuse settings.Results: Compared with \texttt{Fastbot2}, \texttt{CrossBot-With-X} improves activity coverage by 16.5\%--24.2\% and widget coverage by 62.0\%--88.5\% at 60 minutes. It also outperforms a resource-ID-only reuse baseline and sustains coverage growth in four-hour sessions. In the LSH ablation, LSH-backed matching increases throughput by 97.3\% and reduces amortized cost by 46.8\% on average.Conclusion: The results show that widget-level cross-device history reuse can effectively guide phone-target GUI exploration under UI differences, and that efficient approximate matching makes such reuse practical for long-running Android testing.
- Open Access
- 11 June 2026
- Published by Elsevier in SSRN Electronic Journal
SDIC: Distributed Image Compression with Structured Global Context Modeling
Distributed image compression (DIC) reduces redundancy in multi-view systems by exploiting inter-view correlations. However, existing methods struggle to jointly model cross-view dependencies and long-range spatial context efficiently. We propose SDIC, a compact framework for stereo image compression. In the joint decoder, a Cross-view Feature Refinement Module (CFRM) enhances inter-view feature interaction and complementary information utilization. A Structured Global Context Module (SGCM) is further introduced to capture long-range spatial dependencies via multi-directional feature traversal with linear complexity. By combining cross-view refinement and global context modeling, SDIC improves redundancy reduction while maintaining low complexity. Experiments on Cityscapes and InStereo2K demonstrate superior rate–distortion performance, achieving up to 0.8 dB PSNR gain and 1.5% MS-SSIM improvement, while reducing parameters by over 60% and FLOPs by over 75%. Ablation studies validate the effectiveness of both modules.
- 11 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSDIC: Distributed Image Compression with Structured Global Context Modeling
Distributed image compression (DIC) reduces redundancy in multi-view systems by exploiting inter-view correlations. However, existing methods struggle to jointly model cross-view dependencies and long-range spatial context efficiently. We propose SDIC, a compact framework for stereo image compression. In the joint decoder, a Cross-view Feature Refinement Module (CFRM) enhances inter-view feature interaction and complementary information utilization. A Structured Global Context Module (SGCM) is further introduced to capture long-range spatial dependencies via multi-directional feature traversal with linear complexity. By combining cross-view refinement and global context modeling, SDIC improves redundancy reduction while maintaining low complexity. Experiments on Cityscapes and InStereo2K demonstrate superior rate–distortion performance, achieving up to 0.8 dB PSNR gain and 1.5% MS-SSIM improvement, while reducing parameters by over 60% and FLOPs by over 75%. Ablation studies validate the effectiveness of both modules.
- Open Access
- 10 June 2026
- Published by Elsevier in SSRN Electronic Journal
SDIC: Distributed Image Compression with Structured Global Context Modeling
Distributed image compression (DIC) reduces redundancy in multi-view systems by exploiting inter-view correlations. However, existing methods struggle to jointly model cross-view dependencies and long-range spatial context efficiently. We propose SDIC, a compact framework for stereo image compression. In the joint decoder, a Cross-view Feature Refinement Module (CFRM) enhances inter-view feature interaction and complementary information utilization. A Structured Global Context Module (SGCM) is further introduced to capture long-range spatial dependencies via multi-directional feature traversal with linear complexity. By combining cross-view refinement and global context modeling, SDIC improves redundancy reduction while maintaining low complexity. Experiments on Cityscapes and InStereo2K demonstrate superior rate–distortion performance, achieving up to 0.8 dB PSNR gain and 1.5% MS-SSIM improvement, while reducing parameters by over 60% and FLOPs by over 75%. Ablation studies validate the effectiveness of both modules.
- 10 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSDIC: Distributed Image Compression with Structured Global Context Modeling
Distributed image compression (DIC) reduces redundancy in multi-view systems by exploiting inter-view correlations. However, existing methods struggle to jointly model cross-view dependencies and long-range spatial context efficiently. We propose SDIC, a compact framework for stereo image compression. In the joint decoder, a Cross-view Feature Refinement Module (CFRM) enhances inter-view feature interaction and complementary information utilization. A Structured Global Context Module (SGCM) is further introduced to capture long-range spatial dependencies via multi-directional feature traversal with linear complexity. By combining cross-view refinement and global context modeling, SDIC improves redundancy reduction while maintaining low complexity. Experiments on Cityscapes and InStereo2K demonstrate superior rate–distortion performance, achieving up to 0.8 dB PSNR gain and 1.5% MS-SSIM improvement, while reducing parameters by over 60% and FLOPs by over 75%. Ablation studies validate the effectiveness of both modules.
- 9 January 2026
- Published by Springer Nature
- p. 462-476
AMDIC: Adaptive Multi-Granularity Joint Context Transfer for Distributed Image Coding
- 9 January 2026
- Published by Springer Nature
- p. 462-476
AMDIC: Adaptive Multi-Granularity Joint Context Transfer for Distributed Image Coding
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 7032-7047
SHRD: A Scalable Scheme for Hierarchical File Sharing With Rank-Aware Dissemination
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 7032-7047
SHRD: A Scalable Scheme for Hierarchical File Sharing With Rank-Aware Dissemination
- 4 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 12  (2) , 643-660
Generation of Hard SAT Instances and Its Application in Negative Databases for Privacy Enhancement
In recent years, machine learning and deep learning have made remarkable progress and are now widely applied in various fields, including image classification, autonomous driving, natural language processing, and medical diagnosis. However, training these models requires large datasets, which often contain substantial amounts of sensitive personal information, such as medical records and financial details. Without effective privacy protection measures during model training, the risk of sensitive data leakage increases, potentially resulting in severe privacy violations and a loss of trust. As an innovative data representation technique, the Negative Database has proven to be an effective solution in privacy-sensitive domains. Negative databases can be derived from SAT (Boolean Satisfiability Problem) instances, and the hardness of these instances is directly correlated with the level of data protection provided by the Negative Databases. Developing efficient SAT instance generation algorithms to create harder SAT instances can significantly enhance the privacy protection capabilities of the Negative Database. This paper analyzes the hardness conditions of SAT solvers using the Conflict-Driven Clause Learning strategy and proposes a two-stage SAT instance generation algorithm to generate harder SAT instances. These hard instances not only aid in constructing more secure negative databases for enhanced privacy protection but also provide valuable test cases for evaluating and improving SAT solvers.
- 4 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 12  (2) , 643-660
Generation of Hard SAT Instances and Its Application in Negative Databases for Privacy Enhancement
In recent years, machine learning and deep learning have made remarkable progress and are now widely applied in various fields, including image classification, autonomous driving, natural language processing, and medical diagnosis. However, training these models requires large datasets, which often contain substantial amounts of sensitive personal information, such as medical records and financial details. Without effective privacy protection measures during model training, the risk of sensitive data leakage increases, potentially resulting in severe privacy violations and a loss of trust. As an innovative data representation technique, the Negative Database has proven to be an effective solution in privacy-sensitive domains. Negative databases can be derived from SAT (Boolean Satisfiability Problem) instances, and the hardness of these instances is directly correlated with the level of data protection provided by the Negative Databases. Developing efficient SAT instance generation algorithms to create harder SAT instances can significantly enhance the privacy protection capabilities of the Negative Database. This paper analyzes the hardness conditions of SAT solvers using the Conflict-Driven Clause Learning strategy and proposes a two-stage SAT instance generation algorithm to generate harder SAT instances. These hard instances not only aid in constructing more secure negative databases for enhanced privacy protection but also provide valuable test cases for evaluating and improving SAT solvers.
- Open Access
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4821-4823
Poster: LogCADA: Cross-System Log Anomaly Detection based on Two-Stage Multi-Source Domain Adaptation
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4821-4823
Open AccessPoster: LogCADA: Cross-System Log Anomaly Detection based on Two-Stage Multi-Source Domain Adaptation
- Open Access
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4791-4793
Poster: GLog: Self-Evolving Log Anomaly Type Prediction via Instruction-Tuned LLM and Clustering
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4791-4793
Open AccessPoster: GLog: Self-Evolving Log Anomaly Type Prediction via Instruction-Tuned LLM and Clustering
- 21 August 2025
- Published by Springer Nature in Multimedia Systems
- Vol. 31  (5) , 1-13
Jpeg stereo image lossy recompression with mutual information enhancement
- 21 August 2025
- Published by Springer Nature in Multimedia Systems
- Vol. 31  (5) , 1-13
Jpeg stereo image lossy recompression with mutual information enhancement
- Open Access
- 21 August 2025
- Published by Springer Nature in EURASIP Journal on Information Security
- Vol. 2025  (1) , 1-16
Semi-supervised method for anomaly detection in HTTP traffic
- 21 August 2025
- Published by Springer Nature in EURASIP Journal on Information Security
- Vol. 2025  (1) , 1-16
Open AccessSemi-supervised method for anomaly detection in HTTP traffic