A Federated Learning Based Chinese Text Classification Model with Parameter Factorization Weighting

Abstract

Federated learning (FL), as an emerging field of machine learning, has received wide attention since this concept was proposed. In this, paper, we conduct research on text classification based on Federated Learning, and propose a Federated Learning via Local Batch Normalization and Parameter Factorization Weighting based Chinese Text Classification Model (FedBN-PW-CTC). We evaluate our approach on both homogenous and non-homogenous datasets and confirm its effect of 2.95% improvement of accuracy and 4.7% improvement of F1 score on non-homogeneous dataset.