Minutes of Meeting – Differential Fairness Toolkit Project¶
Date: Friday 28th November Time: 4:00 PM Location: Online Present: All four group members (Nick, Kayla, Becky, Raiet)
1. Purpose of the Meeting¶
To discuss early ideas and research for implementing a differential fairness toolkit for the HPDM139 group project.
2. Discussion Summary¶
2.1 Differential Fairness – Approach¶
- The group explored how differential fairness can be implemented in a simple, reusable Python package suitable for health data science workflows.
-
Key considerations included:
-
How to design functions that compute differential fairness metrics.
- How to apply fairness evaluation to typical health datasets (e.g., structured tabular data).
- Making the package easy for other students/health data scientists to import and use.
2.2 Review of Intersectional Fairness Literature¶
- Kayla shared a key research paper on intersectional fairness, explaining the concept of protecting multiple marginalised groups defined by combinations of attributes (e.g., age × sex × ethnicity).
-
The group agreed that intersectional approaches should be included in the toolkit:
-
Handling intersectional subgroups efficiently
- Avoiding small-group instability
- Considering privacy-preserving or smoothed estimators
2.3 Exploration of Reference GitHub Repo¶
- The group examined an existing GitHub repository implementing differential fairness.
-
Noted as a useful reference for:
-
Code structure
- Metric definitions
- Potential API patterns
- Agreed not to copy directly, but to take inspiration for organising our own package.
3. Actions Agreed¶
- All members to read and (try!) to understand the shared intersectional fairness paper in detail.
-
Each member to come back next week with:
-
Ideas on how we might structure our package (modules, functions, workflow).
- Thoughts on which metrics and visualisations to prioritise.
- A firm plan for how we will divide work and deliver the package by the 9th January project deadline.
4. Date of Next Meeting¶
4th December 2025, 10.30am, online.