Research map / Label-efficient and robust learning / Recognition and long-tail
Ensemble Prediction of Task Affinity for Efficient Multi-Task Learning
Afiya Ayman, Ayan Mukhopadhyay, Aron Laszka
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 residual correction stage (ablation) | ETAP · Chemical group gain prediction (|G_train|=10) | Pearson correlation: 0.36 -> 0.50 | Improved |
| add residual correction stage (ablation) | ETAP · CelebA group gain prediction (|G_train|=10) | Pearson correlation: 0.31 -> 0.45 | Improved |
| add residual correction stage (ablation) | ETAP · ETTm1 group gain prediction (|G_train|=10) | Pearson correlation: 0.84 -> 0.85 | No clear effect |
| use B-spline mapping instead of affine mapping (ablation) | ETAP · ETTm1 affinity-to-gain | R2: 0.18 -> 0.68 | Improved |
| use B-spline mapping instead of affine mapping (ablation) | ETAP · CelebA affinity-to-gain | R2: 0.13 -> 0.13 | No clear effect |
| use ETAP affinity score instead of TAG | TAG · Ridership pairwise affinity correlation | Pearson correlation: 0.10 -> 0.36 | Improved |
| use ETAP affinity score instead of TAG | TAG · CelebA affinity runtime | hours: 6.88 -> 2.51 | Improved |
| use ETAP instead of MTGNet with 5 training groups | MTGNet · CelebA group gain prediction | Pearson correlation: 0.10 -> 0.41 | Improved |
| use ETAP instead of Linear Surrogate | Linear Surrogate · ETTm1 transfer sign classification | F1: 0.18 -> 0.31 | Improved |
| use ETAP task grouping instead of PCGrad | PCGrad · ETTm1 | total loss: 4.19 -> 3.89 | Improved |
| use ETAP grouping instead of naive MTL with 4 groups | Naive MTL · Chemical | total loss: 4.83 -> 4.83 | No clear effect |
| apply PCGrad to all tasks | Naive MTL · Chemical | total loss: 4.83 -> 6.20 | Hurt |
Closest papers by meaning
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.88)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.88)
- DMTG: One-Shot Differentiable Multi-Task Grouping (ICML 2024 · similarity 0.86)
- Bayesian Uncertainty for Gradient Aggregation in Multi-Task Learning (ICML 2024 · similarity 0.86)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.86)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (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.
