Research map / Label-efficient and robust learning / Recognition and long-tail
Joint Data-Task Generation for Auxiliary Learning
Hong Chen, Xin Wang, Yuwei Zhou, Yijian Qin, Chaoyu Guan, Wenwu Zhu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add DTG-AuxL generated auxiliary data and task instead of STL | ResNet50 on CUB · CUB | Acc(%): 80.46 -> 81.73 | Improved |
| Add manually collected auxiliary task via JTDS (unhelpful auxiliary) | ResNet50 on CUB · CUB | Acc(%): 80.46 -> 79.34 | Hurt |
| Add DTG-AuxL when the primary data is 5-shot | ResNet50 on CUB · CUB 5-shot | Acc(%): 44.99 -> 52.33 | Improved |
| Add DTG-AuxL when the primary data is 10-shot | ResNet50 on CUB · CUB 10-shot | Acc(%): 63.25 -> 67.68 | Improved |
| Switch backbone from ConvNet to ResNet18 so auxiliary task becomes beneficial | manual auxiliary learning · CIFAR100 | Acc(%): 50.35 -> 49.56 | Mixed |
| Remove label bias term from label generator (ablation) | DTG-AuxL · CUB | Acc(%): 79.28 -> 81.73 | Improved |
| Add label nonlinear term to generator (ablation) | DTG-AuxL · Pet-CUB | Acc(%): 66.58 -> 70.48 | Improved |
| Add feature nonlinear term to generator (ablation) | DTG-AuxL · Pet-CUB | Acc(%): 67.76 -> 70.48 | Improved |
| Add bi-level optimization of the generator (ablation) | DTG-AuxL · Pet-CUB | Acc(%): 68.53 -> 70.48 | Improved |
| Add instance regularization in upper update (ablation) | DTG-AuxL · Pet-CUB | Acc(%): 67.53 -> 70.48 | Improved |
| Add bi-level optimization of the generator (ablation) | DTG-AuxL · Toys | RMSE: 0.9163 -> 0.9153 | No clear effect |
| Replace learnable F-MixUp with joint data-task generator (ablation) | Auxiliary data generation · CUB (ResNet18) | Acc(%): 76.88 -> 77.75 | Improved |
Closest papers by meaning
- Auxiliary Learning with Joint Task and Data Scheduling (ICML 2022 · similarity 0.88)
- AANG : Automating Auxiliary Learning (ICLR 2023 · similarity 0.86)
- Self-Supervised Generalisation with Meta Auxiliary Learning (NeurIPS 2019 · similarity 0.86)
- Aux-NAS: Exploiting Auxiliary Labels with Negligibly Extra Inference Cost (ICLR 2024 · similarity 0.86)
- Auxiliary Learning by Implicit Differentiation (ICLR 2021 · similarity 0.86)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.83)
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.
