Entropy-aided prompt Federated learning

Published: 19 Mar 2024, Last Modified: 03 Apr 2024Tiny Papers @ ICLR 2024 ArchiveEveryoneRevisionsBibTeXCC BY 4.0
Keywords: federated learning, prompt
TL;DR: Entropy-aided prompt Federated learning
Abstract: To overcome the impact of low-quality prompts on the training effect of prompt-Fl, we introduce the information entropy method to calculate the probability distribution of the model output to guide the local client to generate higher-quality prompts. Experimental results show that our method performs well on multiple datasets in both IID and non-IID settings. In addition, our method also leads significantly in training time, showing its efficiency in practical applications.
Submission Number: 21
Loading