AI Paper Atlas

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

SoLar: Sinkhorn Label Refinery for Imbalanced Partial-Label Learning

Haobo Wang, Mingxuan Xia, Yixuan Li, Yuren Mao, Lei Feng, Gang Chen, Junbo Zhao

NeurIPS 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive learning for imbalanced data · 14 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
add Sinkhorn label refinery (prior-matching optimal transport) on top of PRODEN with selection and prior estimation (ablation)PRODEN w S+CPE · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 62.83 -> 74.16Improved
add pre-estimation stage for class prior (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 63.97 -> 74.16Improved
initialize prior from candidate counts instead of pre-estimation (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 74.16 -> 74.34No clear effect
use true class prior (oracle) instead of estimated prior (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 74.16 -> 74.11No clear effect
add class-wise reliable sample selection (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 29.61 -> 74.16Improved
class-wise instead of global sample selection (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 53.01 -> 74.16Improved
add small-loss selection alongside high-confidence selection (ablation)SoLar · CIFAR100-LT (phi=0.1, gamma=20)accuracy (%): 12.80 -> 53.03Improved
add Mixup (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 69.40 -> 74.16Improved
add consistency regularization (ablation)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 57.97 -> 74.16Improved
add test-time logit adjustment with estimated prior (zeta=1)SoLar · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 74.16 -> 80.88Improved
increase logit adjustment strength to zeta=2SoLar-LA · CIFAR10-LT (phi=0.5, gamma=100)accuracy (%): 80.88 -> 72.48Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 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