We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.ST

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Mathematics > Statistics Theory

Title: Bayesian Consistency with the Supremum Metric

Abstract: We present simple conditions for Bayesian consistency in the supremum metric. The key to the technique is a triangle inequality which allows us to explicitly use weak convergence, a consequence of the standard Kullback--Leibler support condition for the prior. A further condition is to ensure that smoothed versions of densities are not too far from the original density, thus dealing with densities which could track the data too closely. A key result of the paper is that we demonstrate supremum consistency using weaker conditions compared to those currently used to secure $\mathbb{L}_1$ consistency.
Comments: 11 pages
Subjects: Statistics Theory (math.ST); Machine Learning (stat.ML)
Cite as: arXiv:2201.03447 [math.ST]
  (or arXiv:2201.03447v1 [math.ST] for this version)

Submission history

From: Nhat Ho [view email]
[v1] Mon, 10 Jan 2022 16:39:25 GMT (14kb)

Link back to: arXiv, form interface, contact.