Teams are the hub’s mechanism for granting fine-grained permissions to access generator models via the API and web interface.

Team’s permissions apply only to users with the role of Team member – all users with administrative or ownership roles will automatically have access to all generators in the organisation.

A Team member who isn't a member of any teams has no access to generator models via the Hub or API.

A Team has two sets of permission controls:

  1. The list of users who are members of the team. These are limited to members of the organisation.

  2. The list of generators that this team grants access to.

Access to a generator can be further limited to only certain models within it, by setting Restrictions. For instance, teams could be limited to only the most private models, by setting a restriction to only access models where epsilon is below a certain threshold.

Creating teams

To create a team select the Access tab from within your organisation, then click “Add new team”.

You can have as many Teams as you have use cases, each with a specific sub-set of generators at appropriate privacy levels.

Assigning generator access permissions

The permission levels you grant here depend on the use case for the team.

If the purpose of the team is to allow access in order to generate high-quality synthetic data for e.g. training other machine learning models, then you can grant access to all models, including non-differentially private ones.

However if the envisioned use case is less quality sensitive, e.g. exploratory data-science work or test-data generation, then you should assign only models with the highest privacy values.

To add a restriction, you must specify:

  • Parameter: The restriction can be based on information from:
    • Model: summary metric scores for privacy, utility and similarity, and the synthesiser the mode was trained on.
    • Training parameters: all training parameter values, such as n_bins, n_parents, epsilon
    • Version: creation date
    • Metadata: Other custom data uploaded with the model, or specified on the version.
  • Condition: Comparison operator to use
  • Value: Value to compare against
  • Compare as: Comparison method, affects equality strictness in some situations.

Within a generator, you can add multiple restrictions, with all restrictions combining additively.

Assigning members

Membership of a team grants read-only access to the specified set of generator models via the API or hub.

Note for roles other than Team member, users automatically have at least read-only access to all generators owned by the organisation, so they are shown in the list of team members, but cannot be removed.