Accepted to 35th USENIX Security Symposium (USENIX Security '26).
A data-free and stealthy server-side backdoor attack against federated LLM-based question-answering systems.
arXiv
BibTeX
@inproceedings{zhu2026aggregator,
title = {When the Aggregator Cheats: Data-Free Backdoors in Federated LLM-based QA Systems},
author = {Zhu, Chenqing and Dai, Yanbo and Tian, Yulong and Li, Qingming and Li, Songze},
booktitle = {Proceedings of the 35th USENIX Security Symposium},
year = {2026},
note = {To appear}
}
Accepted to WASA 2026.
An asynchronous personalized federated learning framework centered on client-side adaptation demands.
Conference Version / arXiv
BibTeX
@inproceedings{zhu2026clientdriven,
title = {Client-Driven Federated Learning for Dynamic Mixtures of Distributions},
author = {Zhu, Chenqing and Li, Songze},
booktitle = {Proceedings of the 21st International Conference on Wireless Algorithms, Systems, and Applications (WASA 2026)},
year = {2026},
note = {To appear}
}
Preprint, arXiv.
A personalized federated continual learning framework that leverages local memory to address client drift and catastrophic forgetting.
arXiv
BibTeX
@article{xie2024fedmes,
title = {FedMeS: Personalized Federated Continual Learning Leveraging Local Memory},
author = {Xie, Jin and Zhu, Chenqing and Li, Songze},
journal = {arXiv preprint arXiv:2404.12710},
year = {2024}
}
