When to Stop Federated Learning: Zero-Shot Generation of Synthetic Validation Data with Generative AI for Early Stopping
Youngjoon Lee, Hyukjoon Lee, Jinu Gong, Yang Cao, Joonhyuk Kang · 2025
Read When to Stop Federated Learning: Zero-Shot Generation of Synthetic Validation Data with Generative AI for Early Stopping by Youngjoon Lee, Hyukjoon Lee, Jinu Gong, Yang Cao, Joonhyuk Kang (2025) 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-2511-11208.txt.
Description
Youngjoon Lee, Hyukjoon Lee, and their colleagues address wasted rounds in federated learning, where many devices train one shared model without pooling private data. Training usually runs for a fixed number of rounds, which wastes compute after the best point and can drag on even when the model fails to learn. A way to know when to stop would save resources.
Their framework uses generative AI to make synthetic validation data with no real examples needed. That fake check set lets the system watch performance and stop near the best round. On chest X-ray classification, it cut training rounds by up to about three quarters while staying within about one percent of best accuracy.
Catalog record
| CREATOR | Youngjoon Lee, Hyukjoon Lee, Jinu Gong, Yang Cao, Joonhyuk Kang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-11208