Research map / Architectures and efficiency / RNNs and normalization
Maximum Roaming 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 |
|---|---|---|---|
| Apply Maximum Roaming instead of fixed parameter partitions (ablation) | Multi-task learning on Celeb-A · Celeb-A | F-score | Helped |
| Use a lower sharing ratio while allowing parameters to roam | Maximum Roaming on Celeb-A · Celeb-A | F-score: 66.23 ± 0.20 vs 65.51 ± 0.32 | Helped |
| Increase the roaming completion rate (ablation) | Maximum Roaming on Celeb-A · Celeb-A | F-score | Helped |
| Use a high update interval Δ (ablation) | Maximum Roaming on Celeb-A · Celeb-A | F-score | Hurt |
| Select update candidates deterministically using parameter cosine similarity instead of uniformly at random (ablation) | Maximum Roaming on Celeb-A · Celeb-A | F-score | Hurt |
| Use roaming rather than fixed masks | Multi-task learning on Cityscapes · Cityscapes | mIoU: 56.52 ± 0.41 -> 57.93 ± 0.20 | Helped |
| Use roaming rather than fixed masks | Multi-task learning on NYUv2 · NYUv2 | mIoU: 16.54 ± 0.02 -> 17.40 ± 0.31 | Helped |
| Use MR instead of MTL | Multi-task learning on NYUv2 · NYUv2 | Mean Err.: 32.43 ± 0.19 -> 30.58 ± 0.04 | Helped |
| Use MR instead of MTL | Multi-task learning on NYUv2 · NYUv2 | Abs. Err.: 60.95 ± 0.41 -> 60.82 ± 0.23 | No clear effect |
| Use MR instead of MTL | Multi-task learning on Cityscapes · Cityscapes | Abs. Err.: 0.0170 ± 0.0006 -> 0.0143 ± 0.0001 | Hurt |
| Reduce the sharing ratio in a fixed partitioning scheme (ablation) | Fixed parameter partitions on Celeb-A · Celeb-A | F-score | Hurt |
Closest papers by meaning
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.86)
- Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · similarity 0.86)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.85)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.85)
- Joint-Task Regularization for Partially Labeled Multi-Task Learning (CVPR 2024 · similarity 0.84)
- DMTG: One-Shot Differentiable Multi-Task Grouping (ICML 2024 · similarity 0.84)
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.