Federated Naive Bayes under Differential Privacy

Towards a Realistic Decentralized Naive Bayes with Differential Privacy

“…While substantial efforts have been devolved to study differentially-private federated learning on deep models, in [16] we argued that, in many practical applications, simpler yet robust models like Naive Bayes classifiers are preferable. Therefore, we proceeded to provide the first (to our knowledge) implementation and evaluation of Federated Naive Bayes with differential privacy, showing that in most cases it can achieve nearly the same performance as a traditional, nondata-private counterpart.…”

“…In this extended version of [16], we provide additional algorithms and experimental results, with the focus on enabling certain realistic scenarios that were not discussed in the original work.…”

“…

This is an extended version of our work in [16]. In this paper, we introduce two novel algorithms to collaboratively train Naive Bayes models across multiple private data sources: Federated Naive Bayes and Gossip Naive Bayes.

…”

“…Accuracy of Federated Naive Bayes for different values of ε and N nodes vs centralized and non-private baselines. Image from[16].…”