Reproduce “Lagrangian Manifold Monte Carlo on Monge Patches” 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

Marcelo Hartmann, Mark Girolami, Arto Klami

Researchers from the University of Helsinki and University of Cambridge developed the Monge metric for geometric Markov Chain Monte Carlo methods, reducing the computational complexity of key operations from cubic to quadratic and broadening its applicability beyond probabilistic models. This advancement improves sampling efficiency and accuracy, particularly for high-dimensional and complex target distributions.

Key insights

  • The Monge metric's structure allows for a significant reduction in computational complexity for metric inverse, determinant, and Christoffel symbol calculations, from cubic to quadratic or linear dependence on dimensionality.
  • The metric intrinsically captures local curvature through its direct relationship to the Hessian of the log-target density, enabling adaptive exploration of complex geometries.
  • The Monge metric is generally applicable to any differentiable target density, removing the model-dependent limitations of the Fisher Information metric.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.