Tilted Mondrian: Triage and Mitigation of Fine-grained Disparities in Conformal Prediction

Key points

• This coarse-to-fine fairness issue raises two questions: how severe can such disparities be, and how can they be mitigated?1
• For severity assessment, we introduce propensity standardization to examine the effect of population composition, and show how group composition can induce disagreement between aggregate and finer-stratum comparisons.1
• For mitigation, we develop Tilted Mondrian to improve fairness compared to group-Mondrian at a finer level without additional model training.1
• Tilted Mondrian retains groupwise coverage guarantees and mitigates coverage and prediction set size disparities within finer strata.1

Sources (1)

• [1]Tilted Mondrian: Triage and Mitigation of Fine-grained Disparities in Conformal Prediction
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 06:36 PM

  Fairness in conformal prediction is often pursued through Mondrian calibration, which provides finite-sample coverage guarantees separately across groups.

  This coarse-to-fine fairness issue raises two questions: how severe can such disparities be, and how can they be mitigated?

Extractive summary: sentences quoted from the sources.