Detecting and Mitigating Backdoor Attacks in OTA-FL Systems: A Two-Stage Robust Aggregation Scheme
Xiaoyan Ma ⋆ , Seohyun Lee ⋆ , Taejoon Kim + , Christopher G. Brinton ⋆ ⋆ Purdue University, + Arizona State University ⋆ {ma946, lee3296, cgb}@purdue.edu, + [email protected] · 2026
Read Detecting and Mitigating Backdoor Attacks in OTA-FL Systems: A Two-Stage Robust Aggregation Scheme by Xiaoyan Ma ⋆ , Seohyun Lee ⋆ , Taejoon Kim + , Christopher G. Brinton ⋆ ⋆ Purdue University, + Arizona State University ⋆ {ma946, lee3296, cgb}@purdue.edu, + [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-19253.txt.
Description
Xiaoyan Ma and three colleagues defend over-the-air federated learning against backdoor attacks, where the server cannot see client updates. Non-IID data makes the task harder, since benign drift can look malicious. Their two-stage framework gives each client a trust score, with indicators chosen by data type and design. The server runs trust-based access, sorting clients into trusted, suspicious, and malicious groups, with suspicious clients facing inspection and reputation checks. Tests on several datasets show stealthy backdoors, including Neurotoxin, are suppressed while accuracy stays competitive.
Catalog record
| CREATOR | Xiaoyan Ma ⋆ , Seohyun Lee ⋆ , Taejoon Kim + , Christopher G. Brinton ⋆ ⋆ Purdue University, + Arizona State University ⋆ {ma946, lee3296, cgb}@purdue.edu, + [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63319 |
| LICENSE | http://creativecommons.org/publicdomain/zero/1.0/ |
| RIGHTS | CC0 |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.19253