Benchmarking federated strategies in Peer-to-Peer Federated learning for biomedical data

TL;DR

This paper studies privacy-preserving biomedical learning by benchmarking multiple aggregation strategies in a fully connected, peer-to-peer Federated Learning (FL) setting across four biomedical datasets. It uses a five-layer dense neural network trained locally on each participant and aggregates parameters via several schemes, including the baseline $ \Phi_j' = \frac{1}{n}\sum_{i=1}^n \Phi_{ji}$ and weighting variants such as $ \Phi_j' = \sum_{i=1}^n \frac{|\mathcal{D}_i|}{\sum_k |\mathcal{D}_k|} \cdot \Phi_{ji}$ and $ \Phi_j' = \sum_{i=1}^n \frac{acc_{ji}}{\sum_k acc_{jk}} \cdot \Phi_{ji}$. It iterates until convergence and evaluates accuracy on held-out test sets as the performance metric. The main finding is that accuracy-based weighting consistently improves federated performance, achieving improvements in 15 of 16 partitions, highlighting the method's robustness and the viability of a decentralized, privacy-preserving approach for biomedical applications.

Abstract

The increasing requirements for data protection and privacy has attracted a huge research interest on distributed artificial intelligence and specifically on federated learning, an emerging machine learning approach that allows the construction of a model between several participants who hold their own private data. In the initial proposal of federated learning the architecture was centralised and the aggregation was done with federated averaging, meaning that a central server will orchestrate the federation using the most straightforward averaging strategy. This research is focused on testing different federated strategies in a peer-to-peer environment. The authors propose various aggregation strategies for federated learning, including weighted averaging aggregation, using different factors and strategies based on participant contribution. The strategies are tested with varying data sizes to identify the most robust ones. This research tests the strategies with several biomedical datasets and the results of the experiments show that the accuracy-based weighted average outperforms the classical federated averaging method.

Figures (2)

• Figure 1: Fully connected peer-to-peer federated learning architecture
• Figure 2: Deep Neural Network topology for the experiments