One-shot federated clustering proposed for non-IID biomedical data
Original title: One-Shot Hierarchical Federated Clustering for Distributed Non-IID Biomedical Data Analysis
Institutions send prototypes once; a server fuses clusterlets to recover global structure while limiting computation and privacy exposure.
The paper proposes a one-shot hierarchical federated clustering framework for multi-institutional biomedical data whose local distributions are non-IID and nested. Clients explore local distributions and send prototypes one way to a server, which aggregates them. A multi-granular competitive learning mechanism fuses clusterlets to limit information loss from that single transmission. The authors say public-dataset experiments outperform state-of-the-art methods while balancing clustering quality, privacy, and computation. No vulnerability or incident is reported.
Unconfirmed: Superiority claims are the authors' and no dataset names or metrics appear in the abstract.
- Communication is one-way prototype transmission from clients to the server.
- The design targets privacy leakage from recursive hierarchical clustering.
- Server-side learning fuses clusterlets after information loss from one-shot transfer.
- Experiments use public datasets and compare state-of-the-art methods.
- No named institutions, datasets, or numeric results are given.
From the original article
1 min read at arxiv.orgIn multi-institutional biomedical data analysis, comprehensive disease profiles are often fragmented across different data silos. Furthermore, local data distributions typically exhibit varying levels of granularity or even complex nested structures.
By Shenghong Cai · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.