Reproduce “Scalable Combinatorial Bayesian Optimization with Tractable Statistical models” 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
Aryan Deshwal, Syrine Belakaria, Janardhan Rao Doppa
A novel method, Parametrized Submodular Relaxation (PSR), addresses scalability and accuracy in combinatorial Bayesian Optimization by replacing semi-definite programming (SDP) with a more efficient submodular relaxation for acquisition function optimization. PSR demonstrates up to 400x speedup and significantly improved acquisition function values compared to SDP, leading to better overall optimization performance.
Key insights
- The acquisition function optimization problem can be reformulated as a parametrized submodular function minimization, allowing for efficient solution.
- Techniques from structured prediction and prescriptive price optimization can be adapted to improve scalability in combinatorial Bayesian Optimization.
- Minimum graph cut algorithms offer a computationally efficient alternative to Semi-Definite Programming for optimizing acquisition functions in combinatorial spaces.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.