RIFLE: Robust Distillation-based FL for Deep Model Deployment on Resource-Constrained IoT Networks
Pouria Arefijamal, Mahdi Ahmadlou, Bardia Safaei, Jörg Henkel, e-mail: {pouria.arefijamal, mahdi.ahmadlou403, bardiasafaei}@sharif.edu, [email protected] · 2026
Read RIFLE: Robust Distillation-based FL for Deep Model Deployment on Resource-Constrained IoT Networks by Pouria Arefijamal, Mahdi Ahmadlou, Bardia Safaei, Jörg Henkel, e-mail: {pouria.arefijamal, mahdi.ahmadlou403, bardiasafaei}@sharif.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-2602-08446.txt.
Description
Tiny internet-of-things devices want to learn together without sharing raw data, but their small models falter on diverse, skewed data and hostile updates. Pouria Arefijamal and three co-authors introduce RIFLE, a federated learning framework that swaps gradient sharing for logit-based knowledge transfer. Their arXiv paper aims to make deep models practical on constrained hardware.
A divergence-based check scores the trustworthiness of each client's update without exposing private data. Tests on three image benchmarks under skewed conditions show false positives down by up to 87.5 percent, poisoning mitigation up by 62.5 percent, and accuracy up by 28.3 percent over standard baselines within ten rounds. Training time for a large vision model collapses from centuries to under two hours on typical devices. Readers get a robust recipe for edge learning.
Catalog record
| CREATOR | Pouria Arefijamal, Mahdi Ahmadlou, Bardia Safaei, Jörg Henkel, e-mail: {pouria.arefijamal, mahdi.ahmadlou403, bardiasafaei}@sharif.edu, [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-08446