Research map / Optimization / Optimizers
FAMO: Fast Adaptive Multitask 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 |
|---|---|---|---|
| replace linear scalarization (average loss) with FAMO adaptive task weighting | multitask model · NYU-v2 | Delta M%: 5.59 -> -4.10 | Helped |
| replace linear scalarization with FAMO | multitask model · QM-9 | Delta M%: 177.6 -> 58.5 | Helped |
| replace linear scalarization with FAMO | multitask model · CelebA | Delta M%: 6.28 -> 4.72 | Helped |
| replace linear scalarization with FAMO | multitask model · CityScapes | Delta M%: 14.11 -> 8.13 | Helped |
| replace NashMTL with FAMO | multitask model · NYU-v2 | Mean rank: 2.11 -> 3.44 | Hurt |
| replace CAGrad with FAMO in multitask RL | SAC · MetaWorld MT10 | success rate: 0.83 -> 0.83 | No clear effect |
| replace PCGrad with FAMO in multitask RL | SAC · MetaWorld MT10 | success rate: 0.72 -> 0.83 | Helped |
| replace NashMTL (authors' reproduction) with FAMO | SAC · MetaWorld MT10 | success rate: 0.80 -> 0.83 | Helped |
| use FAMO instead of NashMTL for training time | toy 2-task problem · toy example | training time: 25x faster | Helped |
| vary regularization gamma over {0.0001, 0.001, 0.01} (ablation) | FAMO · NYU-v2, QM-9, CelebA | Delta M% | No clear effect |
| use larger gamma (0.01 vs smaller) (ablation) | FAMO · CityScapes | Delta M% | Helped |
Closest papers by meaning
- Delve into the Applicability of Advanced Optimizers for Multi-Task Learning (NeurIPS 2026 · similarity 0.87)
- Revisiting Fairness in Multitask Learning: A Performance-Driven Approach for Variance Reduction (CVPR 2025 · similarity 0.85)
- Multi-Task GRPO: Reliable LLM Reasoning Across Tasks (ICML 2026 · similarity 0.84)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.84)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.84)
- AdaMerging: Adaptive Model Merging for Multi-Task Learning (ICLR 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.