Fed-BioMed node: Part 2

Attention

Only start this part after the FL experiment coordinator has told you that your public key has been registered and that the training plans have been sent to your node.

In this step, you will check that your node can reach the researcher, then review and approve the code that will be run on your machine.

Note

All commands assume that the working directory is the root of the brainflex-node repository. They will need to be modified if that is not the case.

Record start time

brainflex record fedbiomed_node_part_2 start

Check VPN connection

Check that the VPN server and the researcher are reachable:

scripts/vpn_status

Example output

** Checking docker containers & VPN access: vpn node
- container vpn is running
- container node is running
- pinging VPN server (10.220.0.1) from container vpn -> OK
- pinging researcher (10.222.0.2) from container vpn -> OK

View and approve the training plans

A Fed-BioMed training plan is the custom Python code that will be run on your node, e.g. for data loading, preprocessing and model definition.

At this point, the Fed-BioMed researcher process should have sent 2 training plans to your node (via the VPN connection), one for each prediction task (brain age regression and biological sex classification). Your node is configured to only run approved code, so you will need to review and approve them before training can start.

Run this to view the training plans (need to run once for each training plan):

docker compose exec -u fedbiomed node fedbiomed node -p /fbm-node training-plan view

This prints the code of the training plan to the terminal.

Once you are satisfied, run this to approve them (also need to run twice):

docker compose exec -u fedbiomed node fedbiomed node -p /fbm-node training-plan approve

Once both training plans are approved, the node is ready for training!

Record end time

brainflex record fedbiomed_node_part_2 end

Share data about setup timing

Timing information and your answers to the initial survey and step-specific questions have been saved to TSV files under the .brainflex directory:

tree .brainflex

Output

.brainflex/
├── initial_survey.tsv
├── step_survey.tsv
└── timestamps.tsv

Take a look at them, make edits as needed, and send them to us via email.

Next

Once all nodes are set up and have approved the training plans, the researcher will launch the federated learning experiment through the network. There is nothing else to do on your side for the moment. Just make sure to:

• Leave the containers running

• Keep the machine turned on and connected to the internet until the end of the experiment (estimated to be 1-2 months)

You can check that the containers are still running with:

docker compose ps

We will keep you updated on the progress of the experiment and manuscript. Thank you for participating!