Reproduce “Feasibility-Driven Trust Region Bayesian Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Paolo Ascia, Elena Raponi, Thomas Bäck, Fabian Duddeck
FuRBO introduces a feasibility-driven trust region Bayesian Optimization algorithm designed to efficiently locate and optimize within sparse, high-dimensional feasible regions. It demonstrates superior performance over existing methods, particularly in severely constrained black-box optimization problems across various dimensions and physics-inspired benchmarks.
Key insights
- Leveraging surrogate model predictions over a sampled neighborhood (inspectors) to define and adapt the trust region, rather than solely relying on previously evaluated points, improves search efficiency.
- Prioritizing the discovery of feasible regions in high-dimensional, sparse constraint landscapes is crucial for practical black-box optimization.
- Dynamically adjusting the trust region's shape and position based on predicted feasibility and optimality allows for more effective navigation of complex, non-convex feasible spaces.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.