Federated Learning with Discriminative Naive Bayes Classifier

TL;DR

The paper tackles privacy-aware federated learning for discrete Naive Bayes classifiers, noting that prior work largely federates probability tables for generative NB and that discriminative NB has been underexplored in this setting. It proposes Federated Weighted NB ($NB^{w}_{fed}$), which federates the discriminative NB by sharing parameter weights $\mathbf{w}$ while keeping the conditional probability tables local, and uses iterative $L$-BFGS-M optimization across clients. The authors evaluate 12 discrete datasets across varying client counts, comparing NB, NB$_{fed}$, NB$^w$, and NB$^{w}_{fed}$, and analyze the effect of limiting inner optimization iterations to mitigate overfitting. Results show that the federated discriminative NB with limited inner iterations achieves strong accuracy and privacy advantages, demonstrating practicality for decentralized classification tasks.

Abstract

Federated Learning has emerged as a promising approach to train machine learning models on decentralized data sources while preserving data privacy. This paper proposes a new federated approach for Naive Bayes (NB) classification, assuming discrete variables. Our approach federates a discriminative variant of NB, sharing meaningless parameters instead of conditional probability tables. Therefore, this process is more reliable against possible attacks. We conduct extensive experiments on 12 datasets to validate the efficacy of our approach, comparing federated and non-federated settings. Additionally, we benchmark our method against the generative variant of NB, which serves as a baseline for comparison. Our experimental results demonstrate the effectiveness of our method in achieving accurate classification.

Figures (3)

• Figure 1: Train accuracy of the algorithms over all the datasets
• Figure 2: Test accuracy of the algorithms over all the datasets
• Figure 3: Mean train and test accuracy