server-side/private-and-fair/README.md

Colabs that generate data for Can a Model Be Differentially Private and Fair?.

• MNIST_Generate_UMAP: Parses privacy rankings and projects MNIST digits with UMAP.
• MNIST DP - Model Grid: Trains a series of differentially private models with different levels of privacy and training data to generate the accuracy tradeoff line charts.
• MNIST DP - Rotated: Trains model with rotated digits for the Subgroup Size and Accuracy Appendix .