AI Paper Atlas

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

Auxiliary Learning with Joint Task and Data Scheduling

Hong Chen, Xin Wang, Chaoyu Guan, Yue Liu, Wenwu Zhu

ICML 2022 · Label-efficient and robust learning › Semi-supervised learning · approach: Self-training theory

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 joint task-data scheduler (JTDS) over naive auxiliary learningauxiliary learning on CUB · CUB fully supervisedAccuracy(%): +2.72 over STL (77.04 vs 73.86)Improved
add joint task-data schedulerauxiliary learning on CUB · CUB semi-supervised (20% labels)Accuracy(%): +12.86 over STL (51.21 vs 38.35)Improved
remove normalized loss term from scheduler (ablation)JTDS · CUB corrupted ratio 0.6Accuracy(%): -6.46 (50.70 to 44.24)Hurt
remove sample feature-prototype term from scheduler (ablation)JTDS · CUB semi-supervised (20% labels)Accuracy(%): -6.58 (51.21 to 44.63)Hurt
restrict scheduler to per-task common weights (only task) (ablation)JTDS · CUB fully supervisedAccuracy(%): -4.14 (72.90 vs 77.04)Hurt
remove auxiliary tasks (train only on target with data scheduling) (ablation)JTDS · CUB fully supervisedAccuracy(%): -1.57 (75.47 vs 77.04)Hurt
naive per-sample learnable weights (N-JTDS) instead of parameter-efficient scheduler (ablation)auxiliary learning on CUB with 312 attribute tasks · CUB fully supervisedAccuracy(%): -5.66 vs JTDS (71.38 vs 77.04)Hurt
joint task-data scheduler vs grid/gradient-based auxiliary weighting on MovieLens-1Mauxiliary learning on MovieLens-1M · MovieLens-1M fully supervisedRMSE: 0.9087 vs 0.9112 STL (lower is better)Improved
add joint task-data schedulerauxiliary learning on CIFAR10 and CIFAR100 fully supervised · CIFAR10 and CIFAR100 fully supervisedAccuracy(%): CIFAR10 +0.99 over STL (72.59 vs 71.60); CIFAR100 +1.54 (75.68 vs 74.14)Mixed
naive per-sample weights (N-JTDS) in supervised pet/CIFAR settingsauxiliary learning with few tasks · Pet fully supervisedAccuracy(%): N-JTDS 67.28 vs STL 61.45Improved

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