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

Recently, semi-supervised federated learning (semi-FL) has been proposed tohandle the commonly seen real-world scenarios with labeled data on the serverand unlabeled data on the clients. However, existing methods face severalchallenges such as communication costs, data heterogeneity, and trainingpressure on client devices. To address these challenges, we introduce thepowerful diffusion models (DM) into semi-FL and propose FedDISC, a FederatedDiffusion-Inspired Semi-supervised Co-training method. Specifically, we firstextract prototypes of the labeled server data and use these prototypes topredict pseudo-labels of the client data. For each category, we compute thecluster centroids and domain-specific representations to signify the semanticand stylistic information of their distributions. After adding noise, theserepresentations are sent back to the server, which uses the pre-trained DM togenerate synthetic datasets complying with the client distributions and train aglobal model on it. With the assistance of vast knowledge within DM, thesynthetic datasets have comparable quality and diversity to the client images,subsequently enabling the training of global models that achieve performanceequivalent to or even surpassing the ceiling of supervised centralizedtraining. FedDISC works within one communication round, does not require anylocal training, and involves very minimal information uploading, greatlyenhancing its practicality. Extensive experiments on three large-scale datasetsdemonstrate that FedDISC effectively addresses the semi-FL problem on non-IIDclients and outperforms the compared SOTA methods. Sufficient visualizationexperiments also illustrate that the synthetic dataset generated by FedDISCexhibits comparable diversity and quality to the original client dataset, witha neglectable possibility of leaking privacy-sensitive information of theclients.

Exploring One-shot Semi-supervised Federated Learning with A Pre-trained Diffusion Model