Reproduce “A General Framework for User-Guided 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
Carl Hvarfner, Frank Hutter, Luigi Nardi
ColaBO, a new framework for Bayesian Optimization, integrates various user beliefs directly into the Gaussian Process surrogate model using a belief-weighted posterior, demonstrating accelerated optimization on synthetic and real-world tasks with accurate priors while maintaining robustness against misleading information.
Key insights
- User beliefs can be integrated in a Bayesian-principled manner by re-weighting the Gaussian Process prior and posterior, effectively biasing the model towards functions consistent with expert knowledge.
- Decoupling prior sampling from the data-dependent posterior update allows for flexible injection of diverse user beliefs without compromising theoretical consistency.
- The framework's design inherently offers robustness to misleading priors, enabling it to recover default BO performance rather than being severely misled.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.