Research map / Learning theory and algorithms / Combinatorial and search optimization
Multi-Objective Population Based Training
Arkadiy Dushatskiy, Alexander Chebykin, Tanja Alderliesten, Peter Bosman
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 |
|---|---|---|---|
| Use domination-based multi-objective ranking rather than scalarization | PBT for multi-objective hyperparameter optimization · Diverse precision/recall, accuracy/fairness, and accuracy/robustness tasks | hypervolume | Improved |
| Use MO-PBT rather than the strongest listed baseline | MO-HPO · Precision/Recall, Adult | hypervolume: 69.61 ± 0.27 -> 70.59 ± 0.13 | Improved |
| Use MO-PBT rather than the strongest listed baseline | MO-HPO · Accuracy/Robustness, CIFAR-10 | hypervolume: 34.74 ± 0.48 -> 35.40 ± 0.09 | Improved |
| Optimize only fairness rather than jointly optimizing the objectives | Single-objective PBT · Accuracy/Fairness tasks | hypervolume | Hurt |
| Use random scalarization instead of domination-based ranking | PBT for multi-objective hyperparameter optimization · Diverse multi-objective tasks | hypervolume | Hurt |
| Use BO-MO-ASHA instead of MO-ASHA | MO-ASHA · Considered multi-objective tasks | hypervolume | No clear effect |
| Use local mutation rather than random mutation (ablation) | MO-PBT explore operator · MO-PBT ablation tasks | hypervolume | Improved |
| Set truncation selection τ to 25 rather than 10 or 50 (ablation) | MO-PBT exploit operator · MO-PBT ablation tasks | hypervolume | Improved |
| Use greedy scattered subset selection rather than crowding distance (ablation) | MO-PBT ranking · MO-PBT ablation tasks | hypervolume | Improved |
| Require accuracy to exceed that of trivial classifiers (ablation) | MO-PBT · CelebA Accuracy/DSP and Accuracy/DSP/DEO | hypervolume | No clear effect |
| Increase population size from 16 to 32 to 64 (ablation) | MO-PBT · Scalability experiments | hypervolume | Improved |
| Expand the image-task search space from 5 variables to 31 variables (ablation) | MO-PBT · Accuracy/Robustness image tasks | hypervolume | Improved |
Closest papers by meaning
- Iterated Population Based Training with Task-Agnostic Restarts (ICML 2026 · similarity 0.89)
- LibMOON: A Gradient-based MultiObjective OptimizatioN Library in PyTorch (NeurIPS 2024 · similarity 0.84)
- Interactive Hyperparameter Optimization in Multi-Objective Problems via Preference Learning (AAAI 2024 · similarity 0.84)
- ParetoFlow: Guided Flows in Multi-Objective Optimization (ICLR 2025 · similarity 0.83)
- Preference-Guided Diffusion for Multi-Objective Offline Optimization (NeurIPS 2025 · similarity 0.82)
- IOP: An Idempotent-Like Optimization Method on the Pareto Front of Hypernetwork (AAAI 2025 · 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.
