Newshot

COSI: A Framework for Robustly Valid Inference

Summary

arXiv:2611.22096v1 Announce Type: cross Abstract: We introduce COSI (COnfidence from Scale Invariance), a framework for constructing confidence sets for statistical functionals. COSI builds on the HulC (Hull-based Confidence), which uses median unbiasedness to construct confidence sets from batch estimates. COSI uses scale-invariance properties shared by many common estimators. We divide the data into batches, compute an estimate in each batch, and retain each candidate parameter value for which the batch estimates pass a test of scale invariance. The resulting confidence sets are asymptotically valid under weaker assumptions than classical Wald and bootstrap procedures require, and their coverage can be closer to the nominal level in small samples. Applications include robust inference for the mean, regression coefficients in misspecified linear models, tensor norms, and Fr\'echet means.

Original text(via arXiv stat.ML)

Statistics > Methodology

Title:COSI: A Framework for Robustly Valid Inference

Abstract:We introduce COSI (COnfidence from Scale Invariance), a framework for constructing confidence sets for statistical functionals. COSI builds on the HulC (Hull-based Confidence), which uses median unbiasedness to construct confidence sets from batch estimates. COSI uses scale-invariance properties shared by many common estimators. We divide the data into batches, compute an estimate in each batch, and retain each candidate parameter value for which the batch estimates pass a test of scale invariance. The resulting confidence sets are asymptotically valid under weaker assumptions than classical Wald and bootstrap procedures require, and their coverage can be closer to the nominal level in small samples. Applications include robust inference for the mean, regression coefficients in misspecified linear models, tensor norms, and Fréchet means.
Comments:Slides based on this paper can be found at this https URL (These slides includes more examples and sometime more simulations. The discussion in some of these slides is also informal.)
Subjects:Methodology (stat.ME) ; Statistics Theory (math.ST); Machine Learning (stat.ML)
Cite as:arXiv:2611.22096 [stat.ME]
(or arXiv:2611.22096v1 [stat.ME] for this version)
https://doi.org/10.48550/arXiv.2611.22096 arXiv-issued DOI via DataCite (pending registration)

Submission history

Access Paper:

  • HTML (experimental)

Current browse context:

References & Citations

  • Google Scholar
  • Semantic Scholar

BibTeX formatted citation

Bookmark

Bibliographic and Citation Tools

Code, Data and Media Associated with this Article

Demos

Recommenders and Search Tools

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by arXiv stat.ML
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗