Fair Regression under Approximate Demographic Parity

Key points

• In regression, this can be too restrictive and reduce predictive accuracy.1
• We instead bound every pairwise Wasserstein distance between group-specific prediction distributions by a chosen tolerance, allowing the groups to differ while keeping fairness tunable.1
• If the transport problem attains this value and suitable Monge maps exist, a deterministic transport-form predictor realizes it.1
• Under squared loss with two positively weighted groups, the $W1$, $W2$, and $W\infty$ constraints yield soft-thresholding, shrinkage, and clipping.1

Sources (1)

• [1]Fair Regression under Approximate Demographic Parity
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 05:05 AM

  Exact demographic parity requires the same prediction distribution across sensitive groups.

  In regression, this can be too restrictive and reduce predictive accuracy.

Extractive summary: sentences quoted from the sources.