Research map / Learning theory and algorithms / Combinatorial and search optimization
Multi-Task Learning as Multi-Objective Optimization
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 MGDA-UB instead of uniform loss scaling | Multi-task LeNet · MultiMNIST | Left digit accuracy (%): 96.46 -> 97.26 | Helped |
| Use MGDA-UB instead of uniform loss scaling | Multi-task LeNet · MultiMNIST | Right digit accuracy (%): 94.99 -> 95.90 | Helped |
| Use MGDA-UB instead of single-task training | Multi-task LeNet · MultiMNIST | Left digit accuracy (%): 97.23 -> 97.26 | No clear effect |
| Use MGDA-UB instead of single-task training | Multi-task LeNet · MultiMNIST | Right digit accuracy (%): 95.90 -> 95.90 | No clear effect |
| Use MGDA-UB instead of uniform loss scaling | Multi-task ResNet-18 · CelebA | Average attribute error: 9.62 -> 8.25 | Helped |
| Use MGDA-UB instead of GradNorm | Multi-task ResNet-18 · CelebA | Average attribute error: 8.44 -> 8.25 | Helped |
| Use MGDA-UB instead of single-task training | Multi-task ResNet-18 · CelebA | Average attribute error: 8.77 -> 8.25 | Helped |
| Use MGDA-UB instead of GradNorm | Multi-task ResNet-50 · Cityscapes validation | Semantic segmentation mIoU (%): 64.81 -> 66.63 | Helped |
| Use MGDA-UB instead of uncertainty weighting | Multi-task ResNet-50 · Cityscapes validation | Instance segmentation error (px): 11.54 -> 10.25 | Helped |
| Use MGDA-UB instead of GradNorm | Multi-task ResNet-50 · Cityscapes validation | Disparity error (px): 2.57 -> 2.54 | Helped |
| Add the MGDA-UB approximation in place of full shared-parameter gradients (ablation) | Multi-task ResNet-50 · Cityscapes validation | Training time: 38.6 -> 23.3 | Helped |
| Add the MGDA-UB approximation in place of full shared-parameter gradients (ablation) | Multi-task ResNet-50 · Cityscapes validation | Semantic segmentation mIoU (%): 66.13 -> 66.63 | Helped |
Closest papers by meaning
- Multi-Task Learning with User Preferences: Gradient Descent with Controlled Ascent in Pareto Optimization (ICML 2020 · similarity 0.91)
- Pareto Multi-Task Learning (NeurIPS 2019 · similarity 0.89)
- Efficiently Identifying Task Groupings for Multi-Task Learning (NeurIPS 2021 · similarity 0.89)
- Pareto Manifold Learning: Tackling multiple tasks via ensembles of single-task models (ICML 2023 · similarity 0.88)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.87)
- Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · similarity 0.87)
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.