AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

RoPAWS: Robust Semi-supervised Representation Learning from Uncurated Data

Sangwoo Mo, Jong-Chyi Su, Chih-Yao Ma, Mido Assran, Ishan Misra, Licheng Yu, Sean Bell

ICLR 2023 · Label-efficient and robust learning › Semi-supervised learning · approach: Holistic semi-supervised pipelines · 2 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
Calibrate target predictions with semi-supervised density modeling and an in-domain priorPAWS · Semi-iNat uncurated, from scratchtest accuracy: 37.7 -> 43.0; +5.3%Improved
Calibrate predictions using RoPAWSPAWS · ImageNet, 1% labelstest accuracy: 66.5 -> 66.9; +0.4%Improved
Calibrate predictions using RoPAWSPAWS · CIFAR-10 + CIFAR-100, 25 labels per classtest accuracy: 78.7 -> 85.6; +6.9%Improved
Calibrate predictions using RoPAWSPAWS · CIFAR-10 + Tiny-ImageNet, 25 labels per classtest accuracy: 60.4 -> 83.1; +22.7%Improved
Add semi-supervised density estimation over labeled and unlabeled samples (ablation)RoPAWS · Semi-iNat uncuratedtest accuracy: 37.7 -> 42.5Improved
Add an in-domain prior to semi-supervised density estimation (ablation)RoPAWS with semi-supervised density · Semi-iNat uncuratedtest accuracy: 42.5 -> 43.0Improved
Add reweighted loss when using the in-domain prior (ablation)RoPAWS with semi-supervised density and in-domain prior · Semi-iNat uncuratedtest accuracy: 42.3 -> 43.0Improved
Use the in-domain prior and reweighted loss without semi-supervised density (ablation)RoPAWS · Semi-iNat uncuratedtest accuracy: 37.7 -> 34.4Hurt
Increase reweight power k to 3.0 (ablation)RoPAWS · Semi-iNat uncuratedtest accuracy: 43.0 -> 41.1Hurt
Increase reweight power k to 5.0 (ablation)RoPAWS · Semi-iNat uncuratedtest accuracy: 43.0 -> 40.1Hurt
Use ImageNet SwAV rather than supervised pre-training (ablation)RoPAWS · Semi-iNat uncurated, pre-trainedtest accuracy: 42.4 -> 44.3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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