Research map / Optimization / Optimizers
SAMO: A Lightweight Sharpness-Aware Approach for Multi-Task Optimization with Joint Global-Local Perturbation
Hao Ban, Gokul Ram Subramani, Kaiyi Ji
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 SAMO | FairGrad · Cityscapes (2-task) | delta_m%: 3.90 -> -0.62 | Improved |
| apply SAMO | FairGrad · NYU-v2 (3-task) | delta_m%: -4.96 -> -6.55 | Improved |
| apply SAMO | FairGrad · CelebA (40 tasks) | delta_m%: 0.37 -> -0.74 | Improved |
| apply SAMO | FairGrad · QM9 (11 tasks) | delta_m%: 57.9 -> 53.0 | Improved |
| apply SAMO | MGDA · Cityscapes (2-task) | delta_m%: 44.14 -> 4.30 | Improved |
| apply SAMO | MGDA · Office-Home (4 tasks) | Avg Acc: 75.24 -> 75.35 | No clear effect |
| apply SAMO | LS · Office-Home (4 tasks) | Avg Acc: 77.27 -> 77.58 | Improved |
| add local information to global SAM (joint perturbation) (ablation) | G-SAM-FairGrad · Cityscapes (2-task) ablation | delta_m%: 0.93 -> -0.62 | Improved |
| add global information to local SAM (joint perturbation) (ablation) | L-SAM-MGDA · Cityscapes (2-task) ablation | delta_m%: 11.94 -> 4.30 | Improved |
| add joint perturbation over global SAM (ablation) | G-SAM-MGDA · CelebA (40 tasks) | delta_m%: 11.78 -> 9.59 | Improved |
| use joint perturbation instead of local SAM (ablation) | L-SAM-MGDA · CelebA (40 tasks) | delta_m%: 8.47 -> 9.59 | Hurt |
| add layerwise normalization of local perturbation (ablation) | SAMO-FairGrad · Cityscapes (2-task) ablation | delta_m%: 0.42 -> -0.62 | Improved |
Closest papers by meaning
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.90)
- Delve into the Applicability of Advanced Optimizers for Multi-Task Learning (NeurIPS 2026 · similarity 0.89)
- Continual Optimization with Symmetry Teleportation for Multi-Task Learning (NeurIPS 2025 · similarity 0.87)
- $\boldsymbol{\mu}\mathbf{P^2}$: Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling (NeurIPS 2024 · similarity 0.87)
- SAMPa: Sharpness-aware Minimization Parallelized (NeurIPS 2024 · similarity 0.87)
- Sharpness-Aware Minimization Efficiently Selects Flatter Minima Late In Training (ICLR 2025 · 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.
