Options

to do the work that�must be done

​

GDPR Art 25: “...appropriate technical and organisational measures…”

myDRE Contracts © 2024 by anDREa BV is licensed under CC BY-NC-SA 4.0

Taking care of data = Trusted Processing Environments

​

Version: 20240330

Remember …

In case your organization is Consortium Lead:

Choose smartly who is:

• Accountable (~Controller)
• Privileged Member (~joined or delegated Controller)�More on myDRE roles: https://support.mydre.org/portal/en/kb/articles/mydreroles

This may make collaboration way easier for:

• Cohort studies
• Federated Learning
• Federated Processing / Analysis
• And it might be possible that a Data Transfer Agreement is not needed

Version: 20240125

Cohort Study Blueprint

Version: 20240330

Federated Learning / Processing Blueprint - I

Institute 1

Institute 2

myDRE

Version: 20240330

Workspace

Gather Data

Workspace

​

• No Data Transfer Agreement needed
• Possibly needed:
• Technology Transfer Agreement
• Data Processing Agreement

Outcome

Gather Data

Workspace

Outcome

Federated Learning / Processing Blueprint - II

Research Institute

Institute 1

myDRE

Create Model

Train Model

Gather Data

Prep Data

Trained Model

​

• No Data Transfer Agreement needed
• Possibly needed:
• Technology Transfer Agreement
• Data Processing Agreement

Version: 20240330

Taking care of data = Trusted Processing Environments

Organisation X

Organisation B

Organisation A

myDRE

Version: 20240330

• Trusted
• Setup time infra�< 30 minutes
• Collaboration

• Trusted
• Setup time infra�> 10 days
• Collaboration

Data Sources

Data Sources

Cohort Studies - Data Gathering

Organisation X

Organisation B

Organisation A

myDRE

Version: 20240330

• Trusted
• Setup time infra�< 30 minutes
• Collaboration

• Trusted
• Setup time infra�> 10 days
• Collaboration

Data Sources

Data Sources

Data Sources

Cohort Studies - Data Issuing

Organisation X

Organisation B

Organisation A

myDRE

Version: 20240330

• Trusted
• Setup time infra�< 30 minutes
• Collaboration

• Trusted
• Setup time infra�> 10 days
• Collaboration

Federated Learning / Processing

Organisation X

Organisation B

Organisation A

myDRE

Version: 20240330

• Trusted
• Setup time infra�< 30 minutes
• Collaboration

• Trusted
• Setup time infra�> 10 days
• Collaboration

Data Sources

Data Sources

Data Sources

Pandemic Readiness

​

Organisation X

Organisation B

Organisation A

myDRE

CDC

Version: 20240330

• Trusted
• Setup time infra�< 30 minutes
• Collaboration

• Trusted
• Setup time infra�> 10 days
• Collaboration

Data Sources

Data Sources

Data Sources

DTA or no DTA? - always consult your privacy lawyer

​

Organisation B

Organisation A

myDRE

Version: 20240330

Data Sources

Data Transfer Agreement Needed?

YES: most likely

NO: most likely - how to proof

YES: if Accountable = Org. B*

MAYBE: if Accountable = Org. A*

NO: if Accountable = Org. A*

MAYBE: if Accountable = Org. B*

* Who is Privileged Member might also impact

Federated Learning with Multiple Workspaces in myDRE

Overview:

• Federated Learning is a machine learning approach where multiple edge devices or servers collaborate to train a model without sharing their local data.
• In this setup, each customer has an Azure Subscription which hosts research Workspaces containing private data.
• An additional Shared Subscription hosts a Federation host and an SMB Share.
• The Federation host is the central node where all the model deltas from different Workspaces are aggregated and evaluated.
• The evolved models are then put in a Private Endpoint linked SMB Share.

Steps:

• Each Workspace uses the latest iteration of the global model to train against its private data.
• The Workspace submits the updated model fragment (delta) to be evaluated together with other participating workspaces' trained model deltas.
• The Federation host aggregates and evaluates the model deltas.
• The evolved model is then put in the Private Endpoint linked SMB Share.
• The Workspaces can then download the evolved model and use it for inference.

Benefits:

• This setup allows multiple Workspaces to collaboratively learn a shared prediction model while keeping all the training data on their original servers, preserving data privacy.
• It also allows customers to leverage the power of Federated Learning without having to manage the underlying infrastructure.

Example:

• A group of hospitals are collaborating to train a model to predict patient outcomes. Each hospital has its own private dataset of patient data, and they do not want to share this data with each other.
• Using Federated Learning, the hospitals can train a shared model without sharing their data. The model is trained in a central location, but the training data never leaves the hospitals.
• This setup allows the hospitals to benefit from the power of machine learning without compromising their patients' privacy.

Version: 20230516

Options to organize - Federated Learning / Processing

issue

Study A

Data Capture

Prep Data for Studies

Study B

Outcome analysis

Outcome analysis

Data Extraction

I

II

III

IV

Overall Agreement

Version: 20230516

Institute Y

Institute X

Study A

Data Capture

Prep Data for Studies

Study B

Outcome analysis

Outcome analysis

Data Extraction

I

II

III

IV

Data Capture

Prep Data for Studies

Data Extraction

I

II

Version: 20230516