Research map / Learning theory and algorithms / Combinatorial and search optimization
PSEO: Optimizing Post-hoc Stacking Ensemble Through Hyperparameter Tuning
Beicheng Xu, Wei Liu, Keyao Ding, Yupeng Lu, Bin Cui
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 |
|---|---|---|---|
| tune ensemble size and diversity weight instead of best fixed pair (ablation) | PSEO base model selection · 80 OpenML tasks | average test rank: 12.3 -> 8.8 | Improved |
| add diversity weight (omega > 0) in subset selection (ablation) | stacking with BQP selection · 80 OpenML tasks | average test rank: 17.8-20.4 (omega=0) -> ~12-14 | Improved |
| use BQP-based selection instead of ALL (ablation) | stacking · 80 OpenML tasks | wins out of 30 configs: 18 of 30 wins | Improved |
| use BQP-based selection instead of BEST (ablation) | stacking · 80 OpenML tasks | wins out of 30 configs: 29 of 30 wins | Improved |
| increase dropout rate from 0 to 0.4 (ablation) | ES blender stacking · 80 OpenML tasks | max weight proportion: decreasing trend (no numbers) | Improved |
| add Retain mechanism to deep stacking (ablation) | multi-layer stacking · 80 OpenML tasks | feature quality across layers | Improved |
| use PSEO instead of AutoGluon- strategy | PSEO · 80 OpenML tasks (VolcanoML pool) | average test rank: 6.19 -> 2.96 | Improved |
| use PSEO with AutoGluon base pool | PSEO · 80 OpenML tasks (AutoGluon pool) | average test rank: 1.90 -> 1.36 | Improved |
| stack all models instead of best per algorithm (ablation) | stacking baselines · 80 OpenML tasks | average test rank: ALL-Linear-L1 6.36 vs BEST-Linear-L1 8.69 | Improved |
| use stacking instead of ensemble selection (ablation) | post-hoc ensemble · 80 OpenML tasks | average test rank: ES 7.78 -> ALL-ES-L2 6.25 | Improved |
| directly search ensemble jointly with models (Autostacker, EO) | one-step ensemble learning · 80 OpenML tasks | average test rank: Single-Best 10.79 vs EO 10.36, Autostacker 9.69 | Hurt |
Closest papers by meaning
- DivBO: Diversity-aware CASH for Ensemble Learning (NeurIPS 2022 · similarity 0.88)
- PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization (AAAI 2025 · similarity 0.84)
- MEPSI: An MDL-Based Ensemble Pruning Approach with Structural Information (AAAI 2024 · similarity 0.84)
- Interactive Hyperparameter Optimization in Multi-Objective Problems via Preference Learning (AAAI 2024 · similarity 0.83)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.83)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 2025 · similarity 0.83)
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.
