Research map / Probabilistic and scientific modeling / Bayesian optimization
Hyperparameter Optimization via Sequential Uniform Designs
Zebin Yang, Aijun Zhang
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 |
|---|---|---|---|
| make search sequential (SeqRand instead of Rand) | random search · 120 HPO tasks (5-fold CV) | tasks won: 106 out of the 120 tasks | Improved |
| make uniform design sequential (SeqUD instead of UD) | uniform design · 120 HPO tasks (5-fold CV) | tasks won: 109 out of the 120 tasks | Improved |
| use uniform design points instead of random in the sequential scheme (SeqUD vs SeqRand) | sequential search · 120 HPO tasks (5-fold CV) | tasks won: 91 (36) | Improved |
| use SeqUD instead of Hyperband | XGBoost HPO · R11 parkinsons tele test RMSE | RMSE: 2.477 -> 2.015 | Improved |
| use SeqUD instead of Hyperband | XGBoost HPO · C20 electricity test accuracy | accuracy: 92.08 -> 92.99 | Improved |
| use SeqUD instead of GP-EI | cliff function · cliff (100 runs) | best value: 0.994 -> 1.000 | No clear effect |
| use SeqUD instead of SeqRand | octopus function · octopus (100 runs) | best value: 2.904 -> 2.996 | Improved |
| use larger per-stage runs n=q=150/250 (1000-run budget) (ablation) | SeqUD · 32 synthetic functions (1000 runs) | function value: 29 out of 32 synthetic functions | Improved |
| use larger per-stage runs n=q=150/250 (ablation) | SeqUD · early runs 100-300 of 1000 | function value | Hurt |
| use UD or Sobol instead of random search or LHS | non-sequential design · 100 runs SVM s=2 | centered discrepancy CD2: 0.003440 -> 0.000035 | Improved |
| use SeqUD instead of AutoSklearn on pipeline optimization | pipeline HPO · 40 data sets | tasks won: 17 out of the 40 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Hyperband — Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization (JMLR 2018) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HPO-B: A Large-Scale Reproducible Benchmark for Black-Box HPO based on OpenML (NeurIPS 2021 · similarity 0.88)
- Frugal Optimization for Cost-related Hyperparameters (AAAI 2021 · similarity 0.88)
- Interactive Hyperparameter Optimization in Multi-Objective Problems via Preference Learning (AAAI 2024 · similarity 0.87)
- Meta-Surrogate Benchmarking for Hyperparameter Optimization (NeurIPS 2019 · similarity 0.87)
- UQ-Guided Hyperparameter Optimization for Iterative Learners (NeurIPS 2024 · similarity 0.85)
- AutoMMLab: Automatically Generating Deployable Models from Language Instructions for Computer Vision Tasks (AAAI 2025 · 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.
