AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

ConfHit: Conformal Generative Design via Nested Testing

Siddhartha Laghuvarapu, Ying Jin, Jimeng Sun

ICLR 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Conformal prediction sets

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use ConfHit nested testing instead of Bonferroni correctionConfHit · SBDD design, alpha=0.1empty set frequency: about 100% -> as low as 16% (N=15)Improved
add nested pruning to certification-only procedure (ablation)ConfHit · SBDD design, alpha=0.1, N=15mean set size: reduces mean set size to about 4 moleculesImproved
add nested pruning to certification-only procedure (ablation)ConfHit · CMO designcertified set size: typically halves the certified set sizeImproved
use max score statistic instead of rank-sum, min, mean, LR (ablation)ConfHit · SBDD TargetDiff, N=5power: 73% of seeds at alpha = 0.1Improved
change conformity score among min, mean, LR, rank-sum, max (ablation)ConfHit · SBDD TargetDiff, N=5error rateNo clear effect
add density-ratio distribution shift adjustment (ablation)ConfHit · SBDD TargetDiff and CMO SelfEditcoverage: without adjustment (w=1) coverage can be violated at stringent error ratesImproved
use a stronger property prediction model instead of a weakened one (ablation)ConfHit · property predictor ablationpowerImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy