AI Paper Atlas

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

Learning Sample Difficulty from Pre-trained Models for Reliable Prediction

Peng Cui, Dan Zhang, Zhijie Deng, Yinpeng Dong, Jun Zhu

NeurIPS 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Confidence calibration objectives · 5 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
replace constant-weight entropy regularization with sample difficulty-aware weighting (ablation)ER · ImageNet1kECE (%): 3.720 -> 1.602Improved
replace constant-weight entropy regularization with sample difficulty-aware weighting (ablation)ER · ImageNet1kTop-1 accuracy (%): 73.68 -> 74.11Improved
add difficulty-aware weighting to cross entropyCE · CIFAR-100ECE (%): 7.251 -> 3.410Improved
PolyLoss baseline vs proposedPolyLoss · ImageNet1kECE (%): 7.882 -> 1.602Improved
use CLIP-ViT-B instead of ViT-B as difficulty scorer (ablation)Proposed · ImageNet1kECE (%): 2.701 -> 1.602Improved
use CLIP-ViT-B instead of MAE-ViT-B as scorer (ablation)Proposed · ImageNet1kECE (%): 1.925 -> 1.602Improved
use CLIP-ViT-B instead of DINOv2 as scorer (ablation)Proposed · ImageNet1kTop-1 accuracy (%): 74.01 -> 74.11Improved
use RMD instead of MD for difficulty (ablation)Proposed · ImageNet1kECE (%): 2.125 -> 1.613Improved
use RMD instead of K-means distance (ablation)Proposed · ImageNet1kTop-1 accuracy (%): 73.78 -> 74.11Improved
use RMD instead of correctness ranking lossCRL · ImageNet1kECE (%): 2.272 -> 1.602Improved
use RMD instead of CRL on CIFAR-10CRL · CIFAR-10ECE (%): 0.957 -> 1.212Hurt
increase alpha from 0.05 to 0.30 in entropy regularization (constant weight) (ablation)ER · ImageNet1kECE (%): 3.308 -> 33.26Hurt

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