AI Paper Atlas

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

Rethinking Safe Semi-supervised Learning: Transferring the Open-set Problem to A Close-set One

Qiankun Ma, Jiyao Gao, Bo Zhan, Yunpeng Guo, Jiliu Zhou, Yan Wang

ICCV 2023 · Label-efficient and robust learning › Semi-supervised learning · approach: Safe and realistic semi-supervised learning · 18 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 consistency regularization loss L_c (ablation)Our method · CIFAR-100 (mismatch ratio 0.3)seen-class accuracy (%): 64.7 -> 70.0Improved
add positive learning loss L_p on high-confidence pseudo labels (ablation)Our method · CIFAR-100 (mismatch ratio 0.3)seen-class accuracy (%): 70.0 -> 71.6Improved
add iterative negative learning loss L_n on low-confidence samples (ablation)Our method · CIFAR-100 (mismatch ratio 0.3)seen-class accuracy (%): 70.0 -> 71.4Improved
combine all three losses L_c, L_p, L_n (ablation)Our method · CIFAR-100 (mismatch ratio 0.3)seen-class accuracy (%): 64.7 -> 72.5Improved
model OOD data as an extra (K+1)th prototype instead of K-way classificationOur method · CIFAR-10 feature distributiont-SNE class separability (qualitative)Improved
close-set (K+1)-way prototype framework versus prior safe SSL methodsT2T / SAFE-STUDENT · CIFAR-100 (mismatch ratio 0.3)seen-class accuracy (%): 69.8 -> 72.5Improved
close-set prototype framework versus OpenMatch at high mismatchOpenMatch · TinyImageNet (mismatch ratio 0.6)seen-class accuracy (%): 37.0 -> 39.9Improved
modify the way of splitting ID and OOD samples (ID&OOD version)Proposed · MNIST unseen-class identificationAUC (%): 95.9 -> 96.4Improved
increase mismatch ratio from 0.3 to 0.6Our-Method · CIFAR-100seen-class accuracy (%): 72.5 -> 70.4Hurt

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