Research map / Learning theory and algorithms / Combinatorial and search optimization
Pareto Multi-Task Learning
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 |
|---|---|---|---|
| Use a set of preference-vector constrained subproblems instead of one balanced multi-task solution | Pareto MTL · MultiMNIST, Multi-Fashion-MNIST, Multi-(Fashion+MNIST), and self-localization | Pareto-solution diversity and overall performance | Helped |
| Add the initialization step that moves each solution toward its assigned feasible preference region (ablation) | Pareto MTL · Synthetic example | Pareto-solution diversity | Helped |
| Add the initialization step that moves each solution toward its assigned feasible preference region (ablation) | Pareto MTL · Multi-FashionMNIST | Pareto-solution diversity | Helped |
| Increase preference vectors from 5 to 25 (ablation) | Pareto MTL · UCI census-income | representation of task trade-offs | Helped |
| Use a very large task-difficulty imbalance (a1 = 50, a2 = 1) rather than low or medium imbalance (ablation) | Pareto MTL · Synthetic multi-objective example | Pareto-solution coverage | Hurt |
| Use evenly distributed preference vectors rather than vectors matched to task difficulty (ablation) | Pareto MTL · Synthetic multi-objective example with highly unbalanced task difficulty | Pareto-solution distribution | Hurt |
| Use task-focused preference vector (0, √2/2, √2/2) instead of balanced (√3/3, √3/3, √3/3) (ablation) | Pareto MTL · UTKFace | gender accuracy: 0.878049 -> 0.885552 | Helped |
| Use task-focused preference vector (√2/2, 0, √2/2) instead of balanced (√3/3, √3/3, √3/3) (ablation) | Pareto MTL · UTKFace | race accuracy: 0.786747 -> 0.792814 | Helped |
| Use task-focused preference vector (√2/2, √2/2, 0) instead of balanced (√3/3, √3/3, √3/3) (ablation) | Pareto MTL · UTKFace | age L1-loss: 13.91812 -> 13.51771 | Helped |
Closest papers by meaning
- Multi-Task Learning with User Preferences: Gradient Descent with Controlled Ascent in Pareto Optimization (ICML 2020 · similarity 0.91)
- Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · similarity 0.85)
- Aligned Multi Objective Optimization (ICML 2025 · similarity 0.84)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.83)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.83)
- Many-Objective Multi-Solution Transport (ICLR 2025 · 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.