AI Paper Atlas

Research map / Label-efficient and robust learning / Noisy labels

Rank-guided Diffusion for Noise Few-Shot Learning

Zelei Wu, Kun Zhou, Xulun Ye, Yifan Mei, Jie Hong, Jieyu Zhao

ICML 2026 · Label-efficient and robust learning › Noisy labels · approach: Contrastive learning on noisy data

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 low-rank guidance to the diffusion refinement (ablation)CRDProto diffusion · MiniImageNet 40% noiseaccuracy: 70.84 -> 80.02Improved
add semantic-weighted prototype on top of low-rank diffusion (ablation)CRDProto · MiniImageNet 40% noiseaccuracy: 80.02 -> 81.39Improved
replace unweighted mean prototype with semantic-weighted prototype (ablation)CRDProto · MiniImageNet 5w5saccuracy: 78.33 -> 90.61Improved
replace unweighted mean prototype with semantic-weighted prototype, severe noise (ablation)CRDProto · Label-swap 40% noiseaccuracy: 70.22 -> 81.39Improved
add semantic-weighted prototype to reduce feature variance (ablation)CRDProto · Ablation feature consistencyintra-class feature variance: 3.29 -> 2.25Improved
add CLIP features with diffusion refinement to ProtoNet (ablation)ProtoNet · Label-swap 40% noiseaccuracy: 57.07 -> 62.05Improved
use ViT-B/32 CLIP backbone instead of ResNet12CRDProto · MiniImageNet 0% label swapaccuracy: 80.25 -> 90.61Improved
use DINO ViT-S backbone instead of CLIP ViT-B/32CRDProto · MiniImageNet 60% label swapaccuracy: 63.77 -> 68.22Improved
use CRDProto (CLIP) instead of ECER-FSL at heavy noiseECER-FSL · MiniImageNet 60% label swapaccuracy: 60.38 -> 63.77Improved
use CRDProto (CLIP) instead of ECER-FSL on clean dataECER-FSL · MiniImageNet 0% label swapaccuracy: 93.26 -> 90.61Hurt
add low-rank guidance (eta > 0) to diffusion, qualitative (ablation)CRDProto · t-SNE feature distributioncluster compactnessImproved

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