FeSViBS: Federated Split Learning of Vision Transformer with Block Sampling

TL;DR

FeSViBS introduces a federated split learning framework with block sampling for vision transformers, improving medical image classification by leveraging intermediate features and pseudo tokens, especially under data-scarce and privacy-sensitive conditions.

Contribution

The paper proposes a novel federated split learning approach with block sampling for vision transformers, enhancing model generalization in medical imaging tasks.

Findings

Outperforms existing SL and FL methods on three datasets.

Effective in both IID and non-IID data settings.

Improves model generalizability with pseudo class tokens.

Abstract

Data scarcity is a significant obstacle hindering the learning of powerful machine learning models in critical healthcare applications. Data-sharing mechanisms among multiple entities (e.g., hospitals) can accelerate model training and yield more accurate predictions. Recently, approaches such as Federated Learning (FL) and Split Learning (SL) have facilitated collaboration without the need to exchange private data. In this work, we propose a framework for medical imaging classification tasks called Federated Split learning of Vision transformer with Block Sampling (FeSViBS). The FeSViBS framework builds upon the existing federated split vision transformer and introduces a block sampling module, which leverages intermediate features extracted by the Vision Transformer (ViT) at the server. This is achieved by sampling features (patch tokens) from an intermediate transformer block and distilling their information content into a pseudo class token before passing them back to the client. These pseudo class tokens serve as an effective feature augmentation strategy and enhances the generalizability of the learned model. We demonstrate the utility of our proposed method compared to other SL and FL approaches on three publicly available medical imaging datasets: HAM1000, BloodMNIST, and Fed-ISIC2019, under both IID and non-IID settings. Code: https://github.com/faresmalik/FeSViBS