01I
Interpretable ZSL
Make zero-shot decisions easier to understand and control through meaningful uncertainty signals and targeted machine unlearning.
- Machine unlearning
- Uncertainty awareness
Zihan Ye AI Researcher at UCAS
Trustworthy Zero-Shot Learning
I am currently an Assistant Researcher at the University of Chinese Academy of Sciences (UCAS), working with Prof. Ling Shao. I received my Ph.D. degree from the Premi Lab of the University of Liverpool in 2025, under the supervision of Prof. Kaizhu Huang. During my undergraduate studies, I worked under the guidance of Prof. Fuyuan Hu. My research centers on trustworthy zero-shot learning, with a particular focus on interpretable zero-shot learning, efficient zero-shot learning, and robust zero-shot learning. I aim to develop learning systems that can recognize unseen concepts in a transparent, data-efficient, and dependable manner.
01
Research agenda
Generalization is only useful when we can trust it.
Zero-shot learning connects visual evidence with semantic knowledge to recognize classes never seen during training. My research asks a harder question: how can that transfer remain understandable, data-efficient, and dependable beyond ideal benchmarks?
02E
Efficient ZSL
Learn reliable visual-semantic correlations from limited seen-class data, reducing annotation and training demands while preserving transfer quality.
- Data-efficient learning
- Test-time adaptation
03R
Robust ZSL
Preserve performance under imperfect data, distribution shifts, imbalanced predictions, and test-time changes encountered in the real world.
- Adversarial robustness
- Class & concept vulnerabilities
Featured research
Segment Anything
Segment Anything with Robust Uncertainty-Accuracy Correlation
Establishing a robust relationship between uncertainty and accuracy so zero-shot segmentation predictions are easier to interpret and assess.
View on ScholarZeroDiff++
Generative Test-Time Adaptation for Zero-Shot Learning
Adapting generative zero-shot models at test time to meet the unknown, changing conditions in which unseen classes actually appear.
View on ScholarAdversarial Robustness
Adversarial Robustness in Zero-Shot Learning: An Empirical Study on Class and Concept-Level Vulnerabilities
An empirical study of adversarial vulnerabilities at both class and concept levels, clarifying where zero-shot systems fail under attack.
View on Scholar
02
Publications
A complete record across zero-shot learning and trustworthy AI.
All publications on Scholar↗CVPRW
↗
Low-Effort Jailbreak Attacks Against Text-to-Image Safety Filters
A. B. Mustafa, Z. Ye, Y. Lu, M. P. Pound, S. N. Gowda
TPAMIEfficient ZSL
↗
ZeroDiff++: Generative Test-Time Adaptation for Zero-Shot Learning
Z. Ye, S. N. Gowda, K. Du, W. Luo, L. Shao
Neurocomputing
↗
A Negative-Anchored Self-Relabeling Strategy for Multi-Label Class-Incremental Learning
K. Du, J. Xie, F. Lyu, Y. Zhou, Z. Ye, G. Liu
ICMLInterpretable ZSL
↗
Segment Anything with Robust Uncertainty-Accuracy Correlation
H. Zhou, M. Toussaint, L. Shao, Z. Ye✉ Corresponding author
arXiv
↗
Compression as an Adversarial Amplifier Through Decision Space Reduction
L. Evans, H. Jandu, Z. Ye, Y. Lu, S. N. Gowda
PR
↗
Negative-Weighted Knowledge Distillation Regularized Graph Convolutional Network for Multi-Label Class-Incremental Learning
K. Du, J. Xie, F. Lyu, Y. Zhou, Z. Ye, W. Li, Y. Li, G. Liu
TIPRobust ZSL
↗
Adversarial Robustness in Zero-Shot Learning: An Empirical Study on Class and Concept-Level Vulnerabilities
Z. Ye, Z. Peng, S. N. Gowda, Y. Yan, H. Xu, L. Shao
ICLREfficient ZSL
↗
ZeroDiff: Solidified Visual-Semantic Correlation in Zero-Shot Learning
Z. Ye, S. N. Gowda, X. Huang, H. Xu, Y. Jin, K. Huang, X. Jin
arXiv
↗
Anyone Can Jailbreak: Prompt-Based Attacks on LLMs and T2Is
A. B. Mustafa, Z. Ye, Y. Lu, M. P. Pound, S. N. Gowda
arXiv
↗
Zero-Shot Robotic Manipulation with Language-Guided Instruction and Formal Task Planning
J. Tang, Z. Ye, Y. Yan, Z. Zheng, T. Gao, Y. Jin
arXiv
↗
Think Small, Plan Smart: Minimalist Symbolic Abstraction and Heuristic Subspace Search for LLM-Guided Task Planning
J. Tang, Y. Yan, Z. Ye, Z. Zheng, Y. Jin
arXiv
↗
DeCLIP: Decoupled Prompting for CLIP-Based Multi-Label Class-Incremental Learning
K. Du, Z. Ye, J. Xie, Y. Shen, Y. Li, F. Hu, L. Shao, G. Liu, J. van de Weijer, et al.
BMVC
↗
Effective Message Hiding with Order-Preserving Mechanisms
Y. Gao, X. Qiu, Z. Ye
TIPRobust ZSL
↗
Rebalanced Zero-Shot Learning
Z. Ye, G. Yang, X. Jin, Y. Liu, K. Huang
PR
↗
Semantic Similarity Distance: Towards Better Text-Image Consistency Metric in Text-to-Image Generation
Z. Tan, X. Yang, Z. Ye, Q. Wang, Y. Yan, A. Nguyen, K. Huang
ACI
↗
A Comprehensive Survey of Zero-Shot Image Classification: Methods, Implementation, and Fair Evaluation
G. Yang, Z. Ye, R. Zhang, K. Huang
TMMInterpretable ZSL
↗
Disentangling Semantic-to-Visual Confusion for Zero-Shot Learning
Z. Ye, F. Hu, F. Lyu, L. Li, K. Huang
AAAI
↗
Multi-Domain Multi-Task Rehearsal for Lifelong Learning
F. Lyu, S. Wang, W. Feng, Z. Ye, F. Hu, S. Wang
ICIP · Oral
↗
Associating Multi-Scale Receptive Fields for Fine-Grained Recognition
Z. Ye, F. Hu, Y. Liu, Z. Xia, F. Lyu, P. Liu
ICME · OralInterpretable ZSL
↗
SR-GAN: Semantic Rectifying Generative Adversarial Network for Zero-Shot Learning
Z. Ye, F. Lyu, L. Li, Q. Fu, J. Ren, F. Hu
Cognitive Computation
↗
Unsupervised Object Transfiguration with Attention
Z. Ye, F. Lyu, L. Li, Y. Sun, Q. Fu, F. Hu
BICS · Best Poster
↗
DAU-GAN: Unsupervised Object Transfiguration via Deep Attention Unit
Z. Ye, F. Lyu, J. Ren, Y. Sun, Q. Fu, F. Hu
03
Academic journey
From representation learning to reliable generalization.
My work connects academic theory with the constraints of real-world AI systems.
Assistant Researcher
University of Chinese Academy of Sciences (UCAS)
Advisor: Prof. Ling ShaoPhD · Computer Science & Engineering
Xi’an Jiaotong-Liverpool University / University of Liverpool
Advisor: Prof. Kaizhu HuangVisiting PhD · TGAI Lab
Westlake University
Supervisor: Prof. Yaochu JinStrategic Intern · Computer Vision
Bosch Corporate Research, Shanghai
04
Service & funding
Supporting research and serving its community.
Leading funded research while contributing to peer review and community organization in machine learning and computer vision.
Review recognition
NeurIPS 2025
Top Reviewer Award · Award rate: 8.02%
Selected funding
Traceable Faithful Zero-Shot Learning
National Natural Science Foundation of China (NSFC) · Young Scientists Fund
- Role
- Principal Investigator
- Budget
- CNY 300,000
- Project period
- 2027–2029
01
Organization
-
GreenMM Workshop · ACM MMOrganizer
02
Peer review
- IEEE Transactions on Multimedia (TMM)Journal Reviewer
- AAAI Conference on Artificial IntelligenceConference Program Committee (Reviewer)
- ICLRConference Reviewer
- CVPRConference Reviewer
- ECCV · ACCVConference Reviewer
- NeurIPSConference Reviewer
- ICCV · ICML · AISTATSConference Reviewer
Research · Collaboration · Ideas