Research map / Optimization / Optimizers
Three-Way Trade-Off in Multi-Objective Learning: Optimization, Generalization and Conflict-Avoidance
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 MoDo instead of MGDA | Multi-objective MNIST MLP · MNIST | Cross-entropy loss: 363.6$\pm$4.1 -> 317.9$\pm$3.4 | Helped |
| Use MGDA instead of static weighting | Multi-objective MNIST MLP · MNIST | Huber loss: 2.2$\pm$0.03 -> 1.9$\pm$0.01 | Helped |
| Use MoDo instead of MGDA | Multi-objective learning · Office-31 | ΔA_id (%): 3.65 ± 0.64 -> 2.26 ± 0.31 | Helped |
| Use MoDo instead of MGDA | Multi-objective learning · Office-home | ΔA_id (%): 5.29 $\pm$ 0.47 -> 2.24 $\pm$ 0.48 | Helped |
| Use MoDo instead of MGDA | Multi-objective learning · NYU-v2 | ΔA_id (%): 6.25$\pm$0.38 -> 3.21$\pm$0.34 | Helped |
| Increase the weight-update step size γ (ablation) | MoDo · Synthetic strongly-convex objectives | CA direction distance | Helped |
| Increase the weight-update step size γ (ablation) | MoDo · Synthetic strongly-convex objectives | PS optimization error | Hurt |
| Increase the weight-update step size γ (ablation) | MoDo · MNIST | CA direction distance | Helped |
| Increase the weight-update step size γ (ablation) | MoDo · MNIST | PS population risk | Hurt |
| Increase the number of iterations T (ablation) | MoDo · MNIST | PS optimization error | Helped |
| Increase the number of iterations T (ablation) | MoDo · MNIST | PS generalization error | Hurt |
| Tune γ between static weighting and MGDA-like updates (ablation) | MoDo · Toy nonconvex objectives | Population risk | Mixed |
Closest papers by meaning
- MGDA Converges under Generalized Smoothness, Provably (ICLR 2025 · similarity 0.87)
- PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization (AAAI 2025 · similarity 0.87)
- Aligned Multi Objective Optimization (ICML 2025 · similarity 0.86)
- Offline Multi-Objective Optimization (ICML 2024 · similarity 0.84)
- Improving Robustness to Multiple Spurious Correlations by Multi-Objective Optimization (ICML 2024 · similarity 0.84)
- FERERO: A Flexible Framework for Preference-Guided Multi-Objective Learning (NeurIPS 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.