Skip to Main content Skip to Navigation
Preprints, Working Papers, ...

PAC-Bayesian aggregation of linear estimators

Abstract : We consider the aggregation of linear estimator in regresison with a sub-Gaussian noise assumption. Aggregating estimators using exponential weights depending on their risk performs well in expectation, but sadly not in probability. A way to overcome this issue is considering exponential weights of a penalized risk. In this case, an oracle inequality can be obtained in probability, but is not sharp. Taking into account the estimated function's norm in the penalty offers a sharp inequality.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

https://hal.inria.fr/hal-01070805
Contributor : Erwan Le Pennec <>
Submitted on : Thursday, October 2, 2014 - 1:35:24 PM
Last modification on : Wednesday, September 16, 2020 - 5:09:16 PM
Long-term archiving on: : Saturday, January 3, 2015 - 10:56:40 AM

Files

pacewa.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01070805, version 1
  • ARXIV : 1410.0661

Collections

Citation

Lucie Montuelle, Erwan Le Pennec. PAC-Bayesian aggregation of linear estimators. 2014. ⟨hal-01070805v1⟩

Share

Metrics

Record views

655

Files downloads

230