Reproduce “Anisotropic Trust Region Warping for Prior-Data Fitted Networks in Bayesian Optimisation” 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
Samuel Younger, Tinkle Chugh, George De Ath
Anisotropic Trust Region (ATR) warping adapts a Prior-Data Fitted Network (PFN) to objective lengthscales without retraining, estimating a scale for each input dimension, rescaling coordinates, and limiting observations when needed. On tested synthetic Matérn objectives, ATR-PFN improves normalized simple regret over PFN at both lengthscale extremes, with a more modest gain for rough functions; broader benchmarks remain untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.