Research map / Probabilistic and scientific modeling / Bayesian optimization
Black-Box Optimization with Local Generative Surrogates
Sergey Shirobokov, Vladislav Belavin, Michael Kagan, Andrei Ustyuzhanin, Atilim Gunes Baydin
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 L-GSO instead of BO | Magnet geometry optimization · Physics experiment | objective value: ~3000 -> ~2200; approximately 25% smaller | Improved |
| Use L-GSO instead of BO | Magnet geometry optimization · Physics experiment | Magnet length (m): 35.44 -> 33.39 | Improved |
| Use L-GSO instead of BO | Magnet geometry optimization · Physics experiment | Magnet weight (kt): 1.27 -> 1.05 | Improved |
| Use L-GSO rather than score function-based optimization | Black-box simulator optimization · Toy experiments | simulator calls to converge: approximately an order of magnitude | Improved |
| Use local generative surrogates on parameters lying on a low-dimensional submanifold | L-GSO · Submanifold optimization problems | simulator calls to converge | Improved |
| Train each local surrogate with fewer parameter points than the full parameter-space dimension (ablation) | L-GSO · Submanifold problems | simulator calls to converge | Improved |
| Reuse previously sampled simulator data in local surrogate training (ablation) | L-GSO · Optimization experiments | simulator calls to attain the minimum | Improved |
| Use BO rather than L-GSO | Low-dimensional Three Hump problem · Three Hump problem | simulator calls to converge | Improved |
| Use BO on a high-curvature objective | Bayesian optimization · Rosenbrock problem | objective value | Hurt |
Closest papers by meaning
- Learning Surrogates for Offline Black-Box Optimization via Gradient Matching (ICML 2024 · similarity 0.88)
- Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization (NeurIPS 2026 · similarity 0.88)
- ROOT: Rethinking Offline Optimization as Distributional Translation via Probabilistic Bridge (NeurIPS 2025 · similarity 0.87)
- Posterior Inference with Diffusion Models for High-dimensional Black-box Optimization (ICML 2025 · similarity 0.87)
- Guided Trajectory Generation with Diffusion Models for Offline Model-based Optimization (NeurIPS 2024 · similarity 0.86)
- Informed Initialization for Bayesian Optimization and Active Learning (NeurIPS 2025 · 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.
