Research map / Learning theory and algorithms / Prediction and decision losses
Iterated Population Based Training with Task-Agnostic Restarts
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 |
|---|---|---|---|
| add data-driven stagnation detection instead of BG-PBT stopping criteria (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| use shrink-perturb instead of copying weights exactly (1.0, 0.0) (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| use shrink-perturb instead of random reinitialization (0.0, 1.0) (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| shrink lambda 0.2 instead of 0.4 or 0.1 (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| add random reinitialization of weights of half the population on restart (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| use time-varying meta BO for HP reinit instead of BG-PBT global BO (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| add random HP sampling for half the population on restart (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| double step size on restart instead of linear increase (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| adjust step size on restart instead of constant step size (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| population multiple 2 instead of 1 or 3 (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
| initial step size 1% instead of 0.5% (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | No clear effect |
| initial step size 1% instead of 3% (ablation) | IPBT · IPBT ablation (4 tasks) | normalized IQM | Improved |
Closest papers by meaning
- Provably Efficient Online Hyperparameter Optimization with Population-Based Bandits (NeurIPS 2020 · similarity 0.89)
- Multi-Objective Population Based Training (ICML 2023 · similarity 0.89)
- Bayesian Optimization for Iterative Learning (NeurIPS 2020 · similarity 0.87)
- Tuning Mixed Input Hyperparameters on the Fly for Efficient Population Based AutoRL (NeurIPS 2021 · similarity 0.86)
- Meta-Surrogate Benchmarking for Hyperparameter Optimization (NeurIPS 2019 · similarity 0.86)
- UQ-Guided Hyperparameter Optimization for Iterative Learners (NeurIPS 2024 · 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.
