Research map / Optimization / Optimizers
Conflict-Averse Gradient Descent for 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 |
|---|---|---|---|
| use CAGrad instead of PCGrad | MTAN · NYU-v2 | delta_m % (lower better): 3.97 -> 0.20 | Helped |
| use CAGrad instead of PCGrad | MTAN · CityScapes | delta_m % (lower better): 18.29 -> 11.64 | Helped |
| use CAGrad instead of MGDA | MTAN · NYU-v2 | delta_m % (lower better): 1.38 -> 0.20 | Helped |
| use CAGrad instead of PCGrad | SAC · Meta-World MT10 | success rate: 0.72 -> 0.83 | Helped |
| use CAGrad instead of PCGrad | SAC · Meta-World MT50 | success rate: 0.50 -> 0.52 | No clear effect |
| sub-sample 4 tasks per step (CAGrad-Fast) (ablation) | CAGrad · Meta-World MT10 | success rate: 0.83 -> 0.82 | No clear effect |
| add CAGrad to ARML auxiliary-task training | ARML · CIFAR-10 500 labels | test accuracy (%): 67.05 -> 68.25 | Helped |
| add MGDA to ARML auxiliary-task training | ARML · CIFAR-10 500 labels | test accuracy (%): 67.05 -> 49.27 | Hurt |
| add CAGrad to GradNorm | GradNorm · CIFAR-10 2000 labels | test accuracy (%): 81.03 -> 83.15 | Helped |
| increase c in CAGrad from 0 toward 10 (ablation) | CAGrad · Multi-Fashion+MNIST | training loss | Mixed |
Closest papers by meaning
- Fair Resource Allocation in Multi-Task Learning (ICML 2024 · similarity 0.87)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.87)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.87)
- PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization (AAAI 2025 · similarity 0.87)
- Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · similarity 0.87)
- Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent (ICML 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.