AI Paper Atlas

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

Deep Discriminative CNN with Temporal Ensembling for Ambiguously-Labeled Image Classification

Yao Yao, Jiehui Deng, Xiuhua Chen, Chen Gong, Jianxin Wu, Jian Yang

AAAI 2020 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 57 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 cross-entropy DCNN with D2CNN loss (entropy + temporal ensembling)ResNet-20 · Lostaccuracy: 0.580 ± 0.031 -> 0.838 ± 0.014Improved
replace cross-entropy DCNN with D2CNN lossResNet-20 · Yahoo!Newsaccuracy: 0.740 ± 0.006 -> 0.833 ± 0.009Improved
replace cross-entropy DCNN with D2CNN lossResNet-20 · SVHN-v3accuracy: 0.908 ± 0.007 -> 0.929 ± 0.001Improved
add discrimination (entropy) term Ld (ablation)D2CNN · LostaccuracyImproved
add temporal ensembling term Lt (ablation)D2CNN · LostaccuracyImproved
ramp weight T(t) up from zero instead of fixed T(t)=Tmax (ablation)D2CNN · LostaccuracyImproved
feed deep features instead of GIST features to baselinesnon-deep baselines (e.g. IPAL) · Lostaccuracy: 0.790 ± 0.034 -> 0.824 ± 0.021Improved
use D2CNN instead of SURE with deep featuresdeep-feature baselines · Lostaccuracy: 0.826 ± 0.006 -> 0.838 ± 0.014Improved
increase false-positive candidate labels from 1 to 3D2CNN · SVHNaccuracy: 0.937 ± 0.003 -> 0.929 ± 0.001No clear effect

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