Feature/async aggregation

[eduard54]

Pull Request

Summary

This pull request introduces CAFF (Cache-based Aggregation with Fairness and Filtering) — a new asynchronous aggregation mechanism for decentralized federated learning (DFL) — into the NEBULA platform.

Key features:

• Implements CAFFUpdateHandler for fully asynchronous model aggregation with update caching and staleness filtering.
• Adds parameters to control dynamic aggregation thresholds, staleness limits, and timeout fallback behavior.
• Integrates CAFF into the frontend, allowing scenario creation with the new mechanism via the UI (only DFL).
• Includes optional code to propagate updates only to peers that contributed in the previous round for CAFF (uncomment code-block in propagator.py marked with comments "TEST CAFF").

This work was developed as part of my bachelor's thesis at the University of Zurich and aims to contribute to NEBULA's asynchronous FL capabilities.

Checklist

• I have read and agree to the Contributor License Agreement.
• I have read and agree to the Contributing Guidelines.
• I have completed the Signed-off-by and Date fields below.

Related Issues

Non explicitly linked - this is a new feature.

Signed-off-by

Signed-off-by: eduard.gash@gmail.com
Date: 2025-08-04