AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Coverage ≠ Exposure: Auditable Control of Same-Support Tail Failures under Multimodal Missingness

Ziteng Hong, Bingzhi Chen, Yishu Liu, Sudong Cai, Guangming Lu

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Prediction-powered and conformal inference

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
add closed-loop TailPressure control (H-CES) over static scheduleStatic-TP · six vision-language routesWorst-Head / Shift-Reweight: Worst-Head +2.1 to 3.2 points; Shift-Reweight +1.4 to 2.5 pointsImproved
add closed-loop TailPressure control (H-CES) over static scheduleStatic-TP · sentiment routes (MulT)Worst-Head / Shift-Reweight: Worst-Head gains of 2.5 to 2.6 points; Shift-Reweight 2.2 to 2.7 pointsImproved
use H-CES instead of LoRA with global weight decayLoRA · MMBench-EN, LLaVA-NeXTWorst-Head: 45.9 -> 53.1Improved
add H-CESFull score · vision-language routesFull: Full changes remain within 0.2 points relative to Static-TPNo clear effect
add closed-loop control (H-CES vs Static-TP), five seedsStatic-TP · MMBench-ENWorst-Head / Shift-Reweight: 2.5 and 2.2 pointsImproved
add closed-loop control (H-CES vs Static-TP), five seedsStatic-TP · MMMUWorst-Head / Shift-Reweight: 2.1 and 1.6 pointsImproved
add train-time gating plus eval-time gating together (full H-CES) (ablation)H-CES · MMBench-ENWorst-Head: 49.1 (train-gated only) vs 53.1; 46.7 (eval-gated only) vs 53.1Improved
keep sample-wise alignment of gates (vs shuffled gates) (ablation)H-CES · MMBench-ENWorst-Head: 46.2 (shuffle train) and 46.0 (shuffle eval) vs 53.1Improved
use correct controller-to-group mapping (vs permuted mapping) (ablation)H-CES · MMBench-ENWorst-Head: 46.9 -> 53.1Improved
use contract-defined groups (vs random grouping of same sizes) (ablation)H-CES · MMBench-ENWorst-Head: 48.2 -> 53.1Improved
use per-group TailPressure control (vs matched global WD schedule or uniform WD budget) (ablation)H-CES · MMBench-ENWorst-Head: 47.6 and 47.9 vs 53.1Improved
use weight-decay actuator (vs per-group LR scaling) (ablation)H-CES · MMBench-ENWorst-Head: 47.4 -> 53.1Improved

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