Dynamic variance-aware federated tuning for efficient autonomous vehicle perception under non-IID settings
V. Dhanavarshini, Sasikumar Periyasamy · 2026
Read Dynamic variance-aware federated tuning for efficient autonomous vehicle perception under non-IID settings by V. Dhanavarshini, Sasikumar Periyasamy (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/PMC13318690.txt.
Description
V. Dhanavarshini and Sasikumar Periyasamy present a 2026 computing study on self-driving car perception. Cars can train shared models without pooling private data through federated learning. Real driving data differ sharply between vehicles, which weakens standard methods. The authors propose a tuning framework that accounts for that variance.
Their framework, DV-FedTune, adjusts each car's contribution dynamically. Weighting reflects update consistency, diversity, and loss-based reliability. It was built around the YOLOv12 detector. Tests used the KITTI object detection dataset under varied uneven-data settings. The method beat three comparison approaches on communication cost, computing cost, and detection performance. Privacy protection also stayed stronger. The authors argue it scales well as more vehicles join.
Catalog record
| CREATOR | V. Dhanavarshini, Sasikumar Periyasamy |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63242 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |