Reproduce “Bayesian Optimization for Min Max 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
Dorina Weichert, Alexander Kister
Researchers at Fraunhofer IAIS adapted information-theoretic Entropy Search (ES) and Knowledge Gradient (KG) acquisition functions for black-box Min Max Optimization. The ES adaptation consistently outperformed a GP-UCB benchmark, converging closer to the true robust optimum on challenging problems like Eggholder*.
Key insights
- Information-centric acquisition functions like ES and KG can effectively balance exploration and exploitation in the complex landscape of Min Max optimization.
- The reformulated integral for ES, relying on products of Heaviside functions, allows for computational tractability when estimating the Min Max point's probability distribution.
- Focusing on gaining information about the optimum's location, rather than simply identifying potentially optimal points, offers an efficient strategy for robust black-box optimization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.