New benchmark shows Mondrian CP improves uncertainty quantification for imbalanced data

A new benchmark study has evaluated various methods for uncertainty quantification in high-stakes decision-making systems, particularly those dealing with imbalanced data and asymmetric error costs. The research found that standard conformal prediction methods significantly under-cover rare, costly minority classes. By contrast, class-conditional (Mondrian) conformal prediction demonstrated a substantial improvement in minority-class coverage. When combined with cost-controlled abstention mechanisms, Mondrian CP effectively reduced expected decision costs compared to traditional approaches, offering practical guidance for deploying reliable decision support systems. AI

IMPACT Improves reliability of AI decision-making in critical applications like healthcare and finance by addressing coverage gaps for rare events.

RANK_REASON The cluster contains an academic paper detailing a new benchmark and methodology for machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Conformal Prediction
• Human-in-the-Loop Abstention
• Mondrian CP

• paper
• safety
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →