Research map / Architectures and efficiency / CNN design
AutoMTL: A Programming Framework for Automating Efficient 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 |
|---|---|---|---|
| add the pre-train stage (ablation) | AutoMTL · CityScapes | overall relative gain Δt (%): +7.5 -> +13.2 | Helped |
| retrain from scratch instead of fine-tune in post-train (ablation) | AutoMTL · CityScapes | overall relative gain Δt (%): +10.1 -> +12.5 | Helped |
| add policy regularization (λreg=0.01) (ablation) | AutoMTL · CityScapes | mIoU: 40.4 -> 43.4 | Helped |
| increase λreg from 0.0005 to 0.01 (ablation) | AutoMTL · CityScapes | params relative (%): -32.3 -> -44.5 | Helped |
| decrease λreg from 0.0005 to 0.0001 (ablation) | AutoMTL · CityScapes | overall relative gain Δt (%): +13.2 -> +9.2 | Hurt |
| λreg=0.001 vs 0.0005 (ablation) | AutoMTL · CityScapes | overall relative gain Δt (%): +12.9 -> +13.2 | No clear effect |
| replace Deeplab-ResNet-34 with MobileNetV2 backbone | AutoMTL · CityScapes | overall relative gain Δt (%): +7.4 | Mixed |
| compare AutoMTL with vanilla Multi-Task on MNasNet | AutoMTL · CityScapes MNasNet | overall relative gain Δt (%): -1.6 -> +9.5 | Helped |
| compare AutoMTL with AdaShare on NYUv2 | AutoMTL · NYUv2 | overall relative gain Δt (%): +3.8 -> +5.3 | Helped |
| compare AutoMTL with AdaShare on Taskonomy | AutoMTL · Taskonomy | overall relative gain Δt (%): +1.2 -> +4.3 | Helped |
| compare AutoMTL with AdaShare on CityScapes | AutoMTL · CityScapes | overall relative gain Δt (%): +10.9 -> +13.2 | Helped |
| compare inference time AutoMTL with independent models | AutoMTL · CityScapes inference time | ms: 71.01 -> 60.84 | Helped |
Closest papers by meaning
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.87)
- Towards Principled Task Grouping for Multi-Task Learning (NeurIPS 2026 · similarity 0.87)
- MTLoRA: Low-Rank Adaptation Approach for Efficient Multi-Task Learning (CVPR 2024 · similarity 0.86)
- Transforming Vision Transformer: Towards Efficient Multi-Task Asynchronous Learner (NeurIPS 2024 · similarity 0.86)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.86)
- Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond (AAAI 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.