[obedjamir]

Fixes #5652.

Description

Adds FedMoSA (MICCAI 2026) as a hosted research example under research/fedmosa: federated SAM adaptation across clients with different imaging modalities, with a custom ModelController for modality-restricted and modality-balanced aggregation and shared-adapter prototype exchange via FLModel.meta, and Client API training. Also adds the project to the research table.

This is an alternative to #5649, which lists the same example as externally maintained; maintainers can choose whichever fits the research folder better.

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Documentation updated.

Testing

• Simulator run with 5 clients (liver: 3 MR-only, 2 CT-only), NVFlare 2.9.0, Python 3.10, NVIDIA H200; results in research/fedmosa/README.md (Validation).
• black, flake8 and isort checks pass locally.

Risk

Self-contained under research/fedmosa; no changes to NVFlare code.

[greptile-apps]

Confidence Score: 5/5

[Low impact] The latest documentation changes appear safe to merge.

Summary

Adds a self-contained FedMoSA research example with modality-specific aggregation and shared prototypes.

• Clients train modality-specific SAM adapters and align shared features.
• The server combines expert updates by modality and balances shared updates.
• Researchers can launch or export the hosted FedMoSA training job.

_{Reviews (12) · Last reviewed commit: "Update validation results" · Reviewed by Greptile}

[greptile-apps]

Want your agent to iterate on Greptile's feedback? Try greploops.