Research map / Optimization / Optimizers
Multi-Objective Deep Learning with Adaptive Reference Vectors
Weiyu Chen, James Kwok
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 |
|---|---|---|---|
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · Multi-MNIST | HV: 2.95 -> 3.02 | Improved |
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · Multi-Fashion+MNIST | HV: 2.82 -> 2.91 | Improved |
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · Adult | HV: 3.336 -> 3.344 | Improved |
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · CelebA-Hard (2 tasks) | HV: 2.217 -> 2.217 | No clear effect |
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · CelebA-Easy (2 tasks) | HV: 3.700 -> 3.710 | Improved |
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · CelebA 5 tasks | HV: 11.56 -> 15.19 | Improved |
| adaptive reference vectors (GMOOAR-U) instead of fixed (COSMOS) | COSMOS · CelebA 15 tasks | HV: 1805 -> 4057 | Improved |
| use HV-based quality function (GMOOAR-HV) instead of uniformity (GMOOAR-U) | GMOOAR-U · CelebA-Hard (2 tasks) | HV: 2.217 -> 2.222 | Improved |
| use HV-based quality function (GMOOAR-HV) instead of uniformity (GMOOAR-U) | GMOOAR-U · Multi-Fashion | HV: 2.33 -> 2.33 | No clear effect |
| use HV-based quality function (GMOOAR-HV) instead of uniformity (GMOOAR-U) | GMOOAR-U · Multi-task learning runtime | runtime: 288 -> 348 | Hurt |
| adaptive RVs overhead relative to COSMOS | COSMOS · Multi-task learning | number of parameters: 43,058 -> 43,685 | No clear effect |
| too small RV learning rate alpha (ablation) | GMOOAR-U · Compass | HV | Hurt |
Closest papers by meaning
- MGDA Converges under Generalized Smoothness, Provably (ICLR 2025 · similarity 0.83)
- PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization (AAAI 2025 · similarity 0.83)
- MAdam: Metric-Aware Multi-Objective Adam (NeurIPS 2026 · similarity 0.83)
- Aligned Multi Objective Optimization (ICML 2025 · similarity 0.82)
- FERERO: A Flexible Framework for Preference-Guided Multi-Objective Learning (NeurIPS 2024 · similarity 0.82)
- Interactive Hyperparameter Optimization in Multi-Objective Problems via Preference Learning (AAAI 2024 · similarity 0.82)
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.
