Thermo-FL: Thermal-Aware Robust Federated Fine-Tuning of Large Language Models for Edge AI
Thermo-FL uses device temperature to decide how much LoRA training and sparse update traffic each edge client should handle.
The paper pairs that client-side throttling with TERRA, a server aggregation pipeline meant to resist corrupted sparse LoRA updates. Its tests use both a large emulator and a Jetson-based physical setup. The authors report stronger BoolQ robustness across clean and attack settings, competitive GSM8K results, stabilized device temperature, smaller compressed uploads, and preserved GSM8K utility under sign-flip/scale and MITM perturbations. ArXiv · AI/CL/LG's note
The paper pairs that client-side throttling with TERRA, a server aggregation pipeline meant to resist corrupted sparse LoRA updates. Its tests use both a large emulator and a Jetson-based physical setup. The authors report stronger BoolQ robustness across clean and attack settings, competitive GSM8K results, stabilized device temperature, smaller compressed uploads, and preserved GSM8K utility under sign-flip/scale and MITM perturbations. ArXiv · AI/CL/LG's note
score 4