English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Multiple medical institutions collaboratively training a model usingfederated learning (FL) has become a promising solution for maximizing thepotential of data-driven models, yet the non-independent and identicallydistributed (non-iid) data in medical images is still an outstanding challengein real-world practice. The feature heterogeneity caused by diverse scanners orprotocols introduces a drift in the learning process, in both local (client)and global (server) optimizations, which harms the convergence as well as modelperformance. Many previous works have attempted to address the non-iid issue bytackling the drift locally or globally, but how to jointly solve the twoessentially coupled drifts is still unclear. In this work, we concentrate onhandling both local and global drifts and introduce a new harmonizing frameworkcalled HarmoFL. First, we propose to mitigate the local update drift bynormalizing amplitudes of images transformed into the frequency domain to mimica unified imaging setting, in order to generate a harmonized feature spaceacross local clients. Second, based on harmonized features, we design a clientweight perturbation guiding each local model to reach a flat optimum, where aneighborhood area of the local optimal solution has a uniformly low loss.Without any extra communication cost, the perturbation assists the global modelto optimize towards a converged optimal solution by aggregating several localflat optima. We have theoretically analyzed the proposed method and empiricallyconducted extensive experiments on three medical image classification andsegmentation tasks, showing that HarmoFL outperforms a set of recentstate-of-the-art methods with promising convergence behavior. Code is availableat https://github.com/med-air/HarmoFL.

HarmoFL: Harmonizing Local and Global Drifts in Federated Learning on Heterogeneous Medical Images