Research map / Learning theory and algorithms / Prediction and decision losses
Asymmetric Multi-task Learning Based on Task Relatedness and Loss
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| asymmetric loss-aware regularization graph (AMTL) vs. single-task learning | AMTL · MNIST / USPS / AWA / Synthetic | error or RMSE | Helped |
| removing task-loss scaling of outgoing edges (AMTL-noLoss) (ablation) | AMTL · MNIST / ImageNet-Room | error: MNIST 14.24 vs 12.92; ImageNet-Room 48.20 vs 40.80 | Hurt |
| symmetric transfer weights (SMTL) instead of asymmetric (ablation) | AMTL · MNIST | error (%): 14.00 vs 12.92 | Hurt |
| curriculum task ordering instead of alternating optimization (ablation) | AMTL · MNIST / USPS / AWA / Synthetic | error or RMSE: MNIST 13.68 vs 12.92; AWA 56.99 vs 56.83; Synthetic 20.33 vs 18.80 | Hurt |
| curriculum task ordering (AMTL-Curriculum) vs. alternating optimization | AMTL · AMTL runtime on synthetic | runtime (s): 7.71 ± 0.16 vs 167.29 ± 7.41 | Helped |
| task-size weighting of outgoing transfer (AMTL-imbalanced) (ablation) | AMTL · ImageNet-Room | prediction error: 40.00 vs 40.80 | Helped |
| task-size weighting of outgoing transfer (AMTL-imbalanced) (ablation) | AMTL · Synthetic (imbalanced) | prediction error: 53.09 vs 53.19 | No clear effect |
| AMTL vs. GO-MTL on School dataset | AMTL · School | RMSE: 10.13 vs 9.87 | Hurt |
| AMTL vs. GO-MTL on ImageNet-Room | AMTL · ImageNet-Room | prediction error: 40.80 vs 47.05 (GO-MTL degrades below STL 45.85) | Helped |
Closest papers by meaning
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.87)
- NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective (NeurIPS 2025 · similarity 0.86)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.86)
- Joint-Task Regularization for Partially Labeled Multi-Task Learning (CVPR 2024 · similarity 0.86)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.86)
- DMTG: One-Shot Differentiable Multi-Task Grouping (ICML 2024 · similarity 0.85)
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.