1. Find the Task and Request to Join

📍 Where you are: Client 1/5 — ① Find · request → ② Workspace · data → ③ Readiness check → ④ Join → ⑤ Wrap-up
Screen: Registry (Console or Studio) · Other side: the Owner approves this request in Owner Track ch. 5

🎯 What you do in this step: Find the Task to join in the Registry, check in the README what the Task does and what data it requires, then request to join. If you join a Task whose data you cannot match, you get blocked at the check in chapter 3. So reading the README before requesting is the Client’s first job.

  1. 1) Find the Task
  2. 2) Read the README — check the conditions to join
  3. 3) Request to join
  4. ⚙️ Explained — the Owner decided how joining works
  5. ✅ Checklist
  6. Next

1) Find the Task

Heart Disease MLP card in the Studio Registry, seen from the Client account

  1. Click Registry in the Agent Studio left menu.
  2. Choose Public Registry.
  3. Click the Heart Disease MLP card. The card title is not the Task name (handson-heart) but the model name. Below the card, the Owner and Task name appear together as <Owner handle> / handson-heart.

Clicking the card opens the Task information on the right. If Participation is approval_required, the Task needs Owner approval. The same card is also visible in Registry in the top menu of the FedOps Console (web).

2) Read the README — check the conditions to join

README on the Task Card — Local data setup

Scroll down the Task information on the right to see the README the Owner wrote, on the Task Card. Check the following before joining.

What to look for in the READMEFor this TaskWhy
Intended useNo disease / disease classification, for educationYou need to know what you are adding your data to
Local data setup (data format)One CSV dataset.csv, 10 columns (age … oldpeak, with cp and restecg limited to fixed values) + target (0/1)Your data must be in this format to join
Federated participationJoining procedure, at least 3 rows per classCheck that your data meets the conditions
LimitationsNot for diagnosis, accuracy around 0.75–0.8 
PrivacyData does not leave the PCCheck what is shared

The Cleveland data in this tutorial comes from the same UCI original, so the format matches. In real life, this is the step where you check whether your hospital’s records match these column names and values.

3) Request to join

Do it in only one place. The result is the same.

WhereHow
FedOps Console (web)Registry → Task card → Federated Learning join button
Agent StudioLeft menu Registry → Public Registry → select the Task → Request to join

This is the request screen in Agent Studio.

Request to join button in Studio

  1. Click Request to join at the top right.

Approval pending state after the request

The button area changes to Approval pending and Withdraw request (cancel the request), and Participation status: pending-approval appears. The card on the left also gets requested. You stay in this waiting state until the Owner approves, and meanwhile you cannot bring the Task into a Workspace.

The screen after approval

When the Owner approves, the card changes to approved and the buttons change to Leave Task · Federated Learning · Open in Workspace. The Overview below also shows Participation status: approved and Completed FL runs: 0. “Complete at least one Federated Learning run after approval before leaving this task.” at the top means you can leave this Task only after completing at least one run.

Tell the Owner you sent the request, and move on to chapter 2 once approved.

⚙️ Explained — the Owner decided how joining works

The Task’s Participation settingWhat you experience
Owner approval required (this tutorial)Request → wait for Owner approval → can join
Open (automatic approval)Can join right after requesting
📚 Federated learning background — what is shared when you join
SharedNot shared
Model parameters trained on my PC (every round)My data file, each individual patient record
The row count of my data (used as the FedAvg weight)Where my data is stored, PC information
Evaluation metrics I allow (accuracy, etc.) 
My account handle, participation record 

Even sending only parameters, attacks that infer some properties of the training data are known to be possible in theory. When joining with sensitive real data, check your institution’s security policy first.

✅ Checklist

  • You found the Task in the Registry
  • You checked that the README’s data format matches your data
  • You requested to join and got the Owner’s approval

Next

2. Workspace and My Data


Back to top

© 2026 Cognitive Computing Lab, Gachon University