A new statistical method called Null-Calibrated Conformal Selection (NCCS) has been proposed, which aims to improve the identification of test candidates with specific response characteristics while controlling false discovery rates. Unlike existing methods that often use prediction-oriented scores, NCCS utilizes target-membership probability as the natural score for selection. This approach is particularly beneficial for complex targets beyond simple mean-monotone cases, such as interval-valued or variance-driven targets, where traditional scores can be misaligned. Experiments demonstrate that NCCS offers finite-sample valid null p-values and can improve selection power on non-mean-monotone targets. AI
IMPACT This method could enhance the precision of identifying relevant data points in AI model training and evaluation, particularly for complex or non-standard target characteristics.
RANK_REASON The item is a research paper published on arXiv detailing a new statistical method. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
