Xiaoyan Zhang
I am a recent graduate from Anhui University. Since Fall 2024, I have been collaborating with the Video and Image Processing Laboratory (VIPER), where I work on continual learning and food recognition. In Fall 2026, I will continue my studies at the University of Michigan, focusing on multi-agent systems, generative models, and large-scale representation learning.
Email: xyzaxis AT umich.edu
Education
Anhui University 2021–2025
Bachelor of Engineering in Artificial Intelligence
GPA: 4.34/5.00, Rank: 1/251
Bachelor of Engineering in Artificial Intelligence
GPA: 4.34/5.00, Rank: 1/251
Selected Publications
SIGMA: Selective-Interleaved Generation with Multi-Attribute Tokens
Xiaoyan Zhang, Zechen Bai, Haofan Wang, Yiren Song
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
Paper / Code /
SIGMA introduces a selective-interleaved generation framework with multi-attribute tokens, enabling more controllable and compositional visual generation across diverse attributes.
Xiaoyan Zhang, Zechen Bai, Haofan Wang, Yiren Song
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
Paper / Code /
@misc{zhang2026sigmaselectiveinterleavedgenerationmultiattribute,
title={SIGMA: Selective-Interleaved Generation with Multi-Attribute Tokens},
author={Xiaoyan Zhang and Zechen Bai and Haofan Wang and Yiren Song},
year={2026},
eprint={2602.07564},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2602.07564},
}SIGMA introduces a selective-interleaved generation framework with multi-attribute tokens, enabling more controllable and compositional visual generation across diverse attributes.
One Adapter for All: Towards Unified Representation in Step-Imbalanced Class-Incremental Learning
Xiaoyan Zhang, Jiangpeng He
arXiv preprint, 2026
Paper / Code /
One-A presents a unified adapter merging framework for step-imbalanced class-incremental learning, improving representation consistency while maintaining efficient deployment with a single merged adapter.
Xiaoyan Zhang, Jiangpeng He
arXiv preprint, 2026
Paper / Code /
@misc{zhang2026adapterallunifiedrepresentation,
title={One Adapter for All: Towards Unified Representation in Step-Imbalanced Class-Incremental Learning},
author={Xiaoyan Zhang and Jiangpeng He},
year={2026},
eprint={2603.10237},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2603.10237},
}One-A presents a unified adapter merging framework for step-imbalanced class-incremental learning, improving representation consistency while maintaining efficient deployment with a single merged adapter.
Teaching Experience
- Fall 2024: ZX52340 Java Technology and Its Application (Practice), Teaching Assistant Anhui University
- Fall 2024: ZJ52014 Introduction to Artificial Intelligence, Teaching Assistant Anhui University
Awards
- Outstanding Graduate of Anhui Province, Ministry of Education of Anhui Province 2025
- National Scholarship, Ministry of Education of the People's Republic of China 2024
- Pacemaker to Merit Student, Anhui University 2022, 2023, 2024
- Merit Student, Anhui University 2022, 2023, 2024
- Song Qingling Future Scholarship (0.05%), The China Song Qingling Foundation 2024
- Excellent Student Scholarship, Anhui University 2023
- Academic Science and Technology Scholarship, Anhui University 2023