Reproduce “Gryffin: An algorithm for Bayesian optimization of categorical variables informed by expert knowledge” 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
Florian Häse, Matteo Aldeghi, Riley J. Hickman, Loïc M. Roch, Alán Aspuru-Guzik
Gryffin introduces a Bayesian optimization framework for efficiently selecting categorical variables in scientific discovery, particularly in chemistry and materials science. It integrates expert physicochemical descriptors to guide the search, with a dynamic variant that learns and refines descriptor relevance, accelerating the discovery of optimal candidates and yielding scientific insights across diverse applications.
Key insights
- Integrating expert domain knowledge through physicochemical descriptors significantly accelerates the optimization of categorical variables, especially in chemical and materials science.
- Dynamically learning and refining descriptor relevance provides deeper scientific insights, identifying the most influential properties for a given target outcome.
- A kernel density estimation approach on the simplex, combined with the concrete distribution and a 'pseudo-boosting' strategy, offers a robust and computationally efficient framework for categorical Bayesian optimization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.