Privacy-Enhanced ZKP-Inspired Framework for Balanced Federated Learning

Abstract

Federated learning (FL) is a distributed machine learning<br> approach that enables remote devices i.e. workers to collaborate to compute<br> the fitting of a neural network model without sharing their data.<br> While this method is favorable to ensure data privacy, an imbalanced<br> data distribution can introduce unfairness in the model training, causing<br> discriminatory bias towards certain under-represented groups. In this paper,<br> we show that imbalance federated data decreases indexes of equity<br> i.e. differences in treatment for underrepresented classes. To address the<br> problem, we propose a federated learning framework called Z-Fed that 1)<br> balances the training without exchange of privacy protected data using<br> a zero knowledge proof (ZKP) technique, and 2) allows for the collection<br> of information on data distributions based on one or more categorical<br> features to produce metadata about population proportions. The proposed<br> framework infers the precise data distribution without exchanging<br> knowledge of the data categories and uses it to coordinate a balanced<br> training set. Z-Fed aims to mitigate the effect of imbalanced data in<br> FL while respecting privacy and without using mediators or probabilistic<br> approaches. Compared to a non-balanced framework, Z-Fed improves<br> fairness and equality measured in equal opportunities (EPD) by 53.54%,<br> equal odds (EOD) by 56.41%, and statistical parity (SPD) by 46.1% on<br> imbalanced UTK datasets, reducing biased predictions among subgroups.<br> EPD, EOD, and SPD measure the disparity of treatment between privileged<br> e.g. over-represented and non-privileged groups. Given the results<br> obtained, Z-Fed can reduce discriminatory behaviors and enhance trustworthy<br> of federated learning.