Skip to main content

Data Owner: Running FL Clients

You retain full control over which Federated Learning (FL) jobs are allowed to access your datasets. This guide walks you through how to review and approve job proposals submitted by Data Scientists.

1. Starting the FL Client​

Once a job is Approved and Run via the RDS-Dashboard, SyftBox automatically initializes the client-side training environment.

  • Secure Sandboxing: SyftBox creates an isolated runtime (via Docker or local process isolation) that contains the approved client_app.py.
  • Automatic Connection: The client automatically connects to the Data Scientist's ServerApp using SyftBox's file sync, requiring no manual port forwarding.

2. Monitoring Training Progress​

You can track the health and status of the training in real-time through the Logs tab:

  • Round Status: The logs will indicate when your client receives a global model and when it starts a local "fit" or "evaluate" round.
  • Performance Metrics: Local loss and accuracy metrics are printed as the model trains on your private data.
  • Resource Usage: Monitor CPU and memory consumption to ensure the job doesn't impact your system's performance.

3. Understanding Model Updates​

  • Gradient Exchange: Your client calculates "updates" (changes to the model's weights) based on your private data.
  • File-Based Sync: In syft-flwr, these updates are saved as temporary files that SyftBox then syncs to the aggregator.
  • Anonymity: The aggregator only receives the mathematical delta of the weights, never the raw rows from your CSV.

Next Step: Proceed to the Privacy & Security guide to understand how your data remains protected throughout the FL process.