eduzhai > Applied Sciences > Engineering >

Siloed Federated Learning for Multi-Centric Histopathology Datasets

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: While federated learning is a promising approach for training deep learningmodels over distributed sensitive datasets, it presents new challenges formachine learning, especially when applied in the medical domain wheremulti-centric data heterogeneity is common. Building on previous domainadaptation works, this paper proposes a novel federated learning approach fordeep learning architectures via the introduction of local-statistic batchnormalization (BN) layers, resulting in collaboratively-trained, yetcenter-specific models. This strategy improves robustness to data heterogeneitywhile also reducing the potential for information leaks by not sharing thecenter-specific layer activation statistics. We benchmark the proposed methodon the classification of tumorous histopathology image patches extracted fromthe Camelyon16 and Camelyon17 datasets. We show that our approach comparesfavorably to previous state-of-the-art methods, especially for transferlearning across datasets.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×