Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning Representation for Multitask Learning through Self-Supervised Auxiliary Learning
Seokwon Shin, Hyungrok Do, Youngdoo Son
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 dummy gradient norm regularization to Vanilla MTL | Vanilla MTL (ResNet50, Split) · UTKFace | Delta_MTL: +2.66% -> +3.50% | Improved |
| add DGR to CAGrad | CAGrad · UTKFace | Delta_MTL: +3.17% -> +4.11% | Improved |
| add DGR to PCGrad | PCGrad · UTKFace | Delta_MTL: +2.80% -> +2.50% | Hurt |
| add DGR to Vanilla MTL | Vanilla MTL (ResNet-50, Split) · NYUv2 | Delta_MTL: +6.32% -> +8.26% | Improved |
| add DGR to PCGrad | PCGrad (Split) · NYUv2 | Delta_MTL: +4.70% -> +7.19% | Improved |
| add DGR to Uncertainty weighting with MTAN | Uncertainty (MTAN) · NYUv2 | Delta_MTL: +7.57% -> +7.28% | Hurt |
| add DGR to IMTL | IMTL · NYUv2 | Delta_MTL: +13.75% -> +14.13% (Split) | Improved |
| add DGR to DWA with MTAN | DWA (MTAN) · CityScapes | Delta_MTL: +4.43% -> +4.31% | Hurt |
| use 3 dummy decoders instead of 1 or 5 (ablation) | DGR · NYUv2 (3 tasks) | Delta_MTL: d=1 +7.13%, d=3 +8.05%, d=5 +7.96% | Improved |
| apply DGR to single-task ImageNet with 5 dummy decoders (ablation) | ResNet-50 · ImageNet | top-1 accuracy: 76.13% -> 75.98% | Hurt |
| add DGR with a Swin backbone | Vanilla MTL (Swin) · NYUv2 | Delta_MTL: +6.74% -> +8.77% | Improved |
| add DGR on Pascal with 5 tasks | Vanilla MTL · Pascal | Delta_MTL: -2.86% -> -2.42% | Improved |
Closest papers by meaning
- The Benefit of Multitask Representation Learning (JMLR 2016 · similarity 0.88)
- Learning from Similar Linear Representations: Adaptivity, Minimaxity, and Robustness (JMLR 2025 · similarity 0.87)
- Joint-Task Regularization for Partially Labeled Multi-Task Learning (CVPR 2024 · similarity 0.87)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.87)
- Improving Few-Shot Learning through Multi-task Representation Learning Theory (ECCV 2022 · similarity 0.87)
- Provable Multi-Task Representation Learning by Two-Layer ReLU Neural Networks (ICML 2024 · similarity 0.86)
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.
