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


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


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

Computer Science > Machine Learning

Title: Meta-Learning PAC-Bayes Priors in Model Averaging

Abstract: Nowadays model uncertainty has become one of the most important problems in both academia and industry. In this paper, we mainly consider the scenario in which we have a common model set used for model averaging instead of selecting a single final model via a model selection procedure to account for this model's uncertainty to improve the reliability and accuracy of inferences. Here one main challenge is to learn the prior over the model set. To tackle this problem, we propose two data-based algorithms to get proper priors for model averaging. One is for meta-learner, the analysts should use historical similar tasks to extract the information about the prior. The other one is for base-learner, a subsampling method is used to deal with the data step by step. Theoretically, an upper bound of risk for our algorithm is presented to guarantee the performance of the worst situation. In practice, both methods perform well in simulations and real data studies, especially with poor-quality data.
Comments: Accepted by AAAI 2020
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
DOI: 10.1609/aaai.v34i04.5841
Cite as: arXiv:1912.11252 [cs.LG]
  (or arXiv:1912.11252v3 [cs.LG] for this version)

Submission history

From: Weiran Huang [view email]
[v1] Tue, 24 Dec 2019 08:55:16 GMT (195kb,D)
[v2] Wed, 25 Dec 2019 03:11:11 GMT (207kb,D)
[v3] Wed, 25 Jan 2023 17:12:14 GMT (196kb,D)

Link back to: arXiv, form interface, contact.