Yu He
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Yu He has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Boheng Li
College of Computing and Data Science, Nanyang Technological University, Nanyang Ave, Singapore
2 shared publications - Mengda Yang
Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, Wuhan, China
2 shared publications - Shuo Shao
Zhejiang University, Hangzhou, China
2 shared publications - Zhan Qin
Zhejiang University
2 shared publications - Dacheng Tao
Nanyang technological university, Singapore, Singapore
1 shared publication Haiyun Zhao
Wuhan Univ. (China)
1 shared publication
Latest Publications
- Open Access
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4642-4656
MIRAGE: Misleading Retrieval-Augmented Generation via Black-box and Query-agnostic Poisoning Attacks
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4642-4656
Open AccessMIRAGE: Misleading Retrieval-Augmented Generation via Black-box and Query-agnostic Poisoning Attacks
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 5864-5879
External Data Extraction Attacks Against Retrieval-Augmented Large Language Models
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 5864-5879
External Data Extraction Attacks Against Retrieval-Augmented Large Language Models
- 2 December 2025
- Published by Elsevier in Neurocomputing
- Vol. 666
Membership inference attacks against split inference via knowledge transfer
- 2 December 2025
- Published by Elsevier in Neurocomputing
- Vol. 666
Membership inference attacks against split inference via knowledge transfer
- Open Access
- 2 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1226-1240
Is Difficulty Calibration All We Need? Towards More Practical Membership Inference Attacks
The vulnerability of machine learning models to Membership Inference Attacks (MIAs) has garnered considerable attention in recent years. These attacks determine whether a data sample belongs to the model's training set or not. Recent research has focused on reference-based attacks, which leverage difficulty calibration with independently trained reference models. While empirical studies have demonstrated its effectiveness, there is a notable gap in our understanding of the circumstances under which it succeeds or fails. In this paper, we take a further step towards a deeper understanding of the role of difficulty calibration. Our observations reveal inherent limitations in calibration methods, leading to the misclassification of non-members and suboptimal performance, particularly on high-loss samples. We further identify that these errors stem from an imperfect sampling of the potential distribution and a strong dependence of membership scores on the model parameters. By shedding light on these issues, we propose RAPID: a query-efficient and computation-efficient MIA that directly Re-leverAges the original membershiP scores to mItigate the errors in Difficulty calibration. Our experimental results, spanning 9 datasets and 5 model architectures, demonstrate that RAPID outperforms previous state-of-the-art attacks (e.g., LiRA and Canary offline) across different metrics while remaining computationally efficient. Our observations and analysis challenge the current de facto paradigm of difficulty calibration in high-precision inference, encouraging greater attention to the persistent risks posed by MIAs in more practical scenarios.
- 2 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1226-1240
Open AccessIs Difficulty Calibration All We Need? Towards More Practical Membership Inference Attacks
The vulnerability of machine learning models to Membership Inference Attacks (MIAs) has garnered considerable attention in recent years. These attacks determine whether a data sample belongs to the model's training set or not. Recent research has focused on reference-based attacks, which leverage difficulty calibration with independently trained reference models. While empirical studies have demonstrated its effectiveness, there is a notable gap in our understanding of the circumstances under which it succeeds or fails. In this paper, we take a further step towards a deeper understanding of the role of difficulty calibration. Our observations reveal inherent limitations in calibration methods, leading to the misclassification of non-members and suboptimal performance, particularly on high-loss samples. We further identify that these errors stem from an imperfect sampling of the potential distribution and a strong dependence of membership scores on the model parameters. By shedding light on these issues, we propose RAPID: a query-efficient and computation-efficient MIA that directly Re-leverAges the original membershiP scores to mItigate the errors in Difficulty calibration. Our experimental results, spanning 9 datasets and 5 model architectures, demonstrate that RAPID outperforms previous state-of-the-art attacks (e.g., LiRA and Canary offline) across different metrics while remaining computationally efficient. Our observations and analysis challenge the current de facto paradigm of difficulty calibration in high-precision inference, encouraging greater attention to the persistent risks posed by MIAs in more practical scenarios.
- 31 October 2009
- Published by SPIE-Intl Soc Optical Eng
- Vol. 7146, 714620
The analysis of land use spatial patterns responded to different geomorphology type: two case studies in Hubei Province, China
Land use spatial pattern reflects both the impact of human activities on the process of ecological system and the human activities response to natural condition. Geomorphology condition, which has a close relationship with land use spatial pattern, is a direct and key factor influencing human to use land at the local level. This paper analyses two cases of land use spatial patterns in Hubei province China, Zigui County and Jiayu County, which are exemplary for two different geomorphology conditions, respectively the mountain area and plain model. The land use spatial pattern is analyzed using geographical information system (ArcGIS) and landscape-pattern analysis software (FRAGSTATS 3.3). The results indicate that different land use type has different landscape characteristics. Simultaneity, the same land use type forms different landscape pattern in the dissimilar geomorphology region. Terrain feature, which restrains the possible land use type, is the base for people to decide how to use the land. It is the most direct factor influencing people behavior for land using. Cultivated land is the substratum of landscape class in Plain, while in Mountain area is forestland. Patches of different landscape classes are inlaid the substratum. Comparing with mountain area, there are fewer land use types and inconspicuous substratum of landscape class in plain because of human intensive disturbance. Meanwhile patches in plain possess higher geometrical similarity and lower landscape fragment than in mountain area.
- 31 October 2009
- Published by SPIE-Intl Soc Optical Eng
- Vol. 7146, 714620
The analysis of land use spatial patterns responded to different geomorphology type: two case studies in Hubei Province, China
Land use spatial pattern reflects both the impact of human activities on the process of ecological system and the human activities response to natural condition. Geomorphology condition, which has a close relationship with land use spatial pattern, is a direct and key factor influencing human to use land at the local level. This paper analyses two cases of land use spatial patterns in Hubei province China, Zigui County and Jiayu County, which are exemplary for two different geomorphology conditions, respectively the mountain area and plain model. The land use spatial pattern is analyzed using geographical information system (ArcGIS) and landscape-pattern analysis software (FRAGSTATS 3.3). The results indicate that different land use type has different landscape characteristics. Simultaneity, the same land use type forms different landscape pattern in the dissimilar geomorphology region. Terrain feature, which restrains the possible land use type, is the base for people to decide how to use the land. It is the most direct factor influencing people behavior for land using. Cultivated land is the substratum of landscape class in Plain, while in Mountain area is forestland. Patches of different landscape classes are inlaid the substratum. Comparing with mountain area, there are fewer land use types and inconspicuous substratum of landscape class in plain because of human intensive disturbance. Meanwhile patches in plain possess higher geometrical similarity and lower landscape fragment than in mountain area.