AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Predictive Inference with Feature Conformal Prediction

Jiaye Teng, Chuan Wen, Dinghuai Zhang, Yoshua Bengio, Yang Gao, Yang Yuan

ICLR 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 3 citations (OpenAlex)

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
Feature CP (feature-space conformal prediction) instead of vanilla CP in output spaceSynthetic multi-dimensional regression · Synthetic multi-dim regression (alpha=0.1)band length: 0.401 -> 0.373Improved
Feature CP instead of vanilla CP in output spaceCityscapes segmentation · Cityscapes segmentation (alpha=0.1)band length: 40.15 -> 1.032Improved
Feature CP instead of vanilla CP in output spaceSynthetic multi-dimensional regression · Synthetic multi-dim regression (alpha=0.1)coverage: 89.91 -> 90.13No clear effect
Feature CP instead of vanilla CP in output spaceCityscapes segmentation · Cityscapes segmentation (alpha=0.1)coverage: 91.41 -> 90.77Mixed
Feature CP applied to pretrained ImageNet classifiers (vs RAPS/APS)ImageNet classification, ResNet 18/50/101, ResNeXt 101, VGG 16 · ImageNet (alpha=0.1)average set length: e.g., ResNet 18 4.43 -> 3.82 vs RAPSImproved
Feature CP applied to ShuffleNet on ImageNetShuffleNet · ImageNet (alpha=0.1)average set length: 5.05 (RAPS) vs 5.06 (Feature CP)No clear effect
Random (unlearned) network features used for Feature CP instead of trained semantic features (ablation)Feature CP · Feature CP vs vanilla CP with same networkband lengthNo clear effect
Feature CQR (feature-level version of CQR) instead of CQRCQR on UCI and other 1-D regression datasets · 1-D regression benchmarks (alpha=0.1)interval lengthImproved
Splitting point choice for feature layer (ablation)Feature CP on small and large networks · Coverage robustnesscoverageNo clear effect

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