Research map / Probabilistic and scientific modeling / Bayesian optimization
GIT-BO: High-Dimensional Bayesian Optimization with Tabular Foundation Models
Rosen Ting-Ying Yu, Cyril Picard, Faez Ahmed
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 adaptive 92.5% variance-explained subspace instead of fixed r=10 (ablation) | GIT-BO · GI-subspace dimension sweep (4 problems) | average rank: 5.5 -> 1.75 | Improved |
| increase subspace dimension to r=40 (ablation) | GIT-BO · GI-subspace dimension sweep (4 problems) | average rank: 5.5 -> 8.0 | Hurt |
| use r=5 instead of r=10 (ablation) | GIT-BO · GI-subspace dimension sweep (4 problems) | average rank: 5.5 -> 3.25 | Improved |
| lower UCB beta from 2.32 (S=512) to 1.65 (q=95%) (ablation) | GIT-BO · UCB beta sweep | average rank: 3.25 -> 2.0 | Improved |
| raise UCB beta to 2.45 (S=1024) (ablation) | GIT-BO · UCB beta sweep | average rank: 3.25 -> 4.75 | Hurt |
| add GI-subspace instead of trust region or BAxUS projection (ablation) | TabPFN BO · high-dimensional BO benchmarks | final regret | Improved |
| add trust region or BAxUS projection to vanilla TabPFN BO (ablation) | TabPFN BO · high-dimensional BO benchmarks | regret | Improved |
| use UCB instead of EI (ablation) | GIT-BO · five BO problems | regret | Improved |
| use centroid of observed data instead of incumbent as x_ref (ablation) | GIT-BO · six BO benchmarks | regret | Improved |
| use random or Sobol instead of uniform sampling in GI-subspace (ablation) | GIT-BO · selected BO benchmarks | regret | Mixed |
| raise initial sample size from 20 to 1000 (ablation) | GIT-BO · five BO problems | statistical rank | No clear effect |
| use GIT-BO instead of GP baselines at large initial sample size N_init=1000 (ablation) | GIT-BO vs GP-based BO · five BO problems | statistical rank | Improved |
Closest papers by meaning
- BOIDS: High-Dimensional Bayesian Optimization via Incumbent-Guided Direction Lines and Subspace Embeddings (AAAI 2025 · similarity 0.89)
- CoffeeBoost: Gradient Boosting Native Conformal Inference for Bayesian Optimization (AAAI 2025 · similarity 0.89)
- Understanding High-Dimensional Bayesian Optimization (ICML 2025 · similarity 0.89)
- Informed Initialization for Bayesian Optimization and Active Learning (NeurIPS 2025 · similarity 0.88)
- Vanilla Bayesian Optimization Performs Great in High Dimensions (ICML 2024 · similarity 0.88)
- BONSAI: Bayesian Optimization with Natural Simplicity and Interpretability (NeurIPS 2026 · similarity 0.88)
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.
