Create an aggregated embedding for membership inference hot spots #41
Labels
metrics visualization
Part of the three phase metrics visualization project
needs validation
Some validation is required to document the solution
Is your feature request related to a problem? Please describe.
We need a clarifying visual to showcase the privacy risks of membership inference, especially as it varies between classes.
Describe the solution you'd like.
An aggregated embedding similar to Figure 4 of Understanding and Visualizing Data Iteration in Machine Learning would be useful. Highlight the worst case.
Describe alternatives you've considered.
Please comment any alternatives.
The text was updated successfully, but these errors were encountered: