Improved Risk Tail Bounds for On-Line Algorithms

Abstract

We prove the strongest known bound for the risk of hypotheses selected from the ensemble generated by running a learning algorithm incremen(cid:173) tally on the training data. Our result is based on proof techniques that are remarkably different from the standard risk analysis based on uniform convergence arguments.

Cite

Text

Cesa-bianchi and Gentile. "Improved Risk Tail Bounds for On-Line Algorithms." Neural Information Processing Systems, 2005.

Markdown

[Cesa-bianchi and Gentile. "Improved Risk Tail Bounds for On-Line Algorithms." Neural Information Processing Systems, 2005.](https://mlanthology.org/neurips/2005/cesabianchi2005neurips-improved/)

BibTeX

@inproceedings{cesabianchi2005neurips-improved,
  title     = {{Improved Risk Tail Bounds for On-Line Algorithms}},
  author    = {Cesa-bianchi, Nicolò and Gentile, Claudio},
  booktitle = {Neural Information Processing Systems},
  year      = {2005},
  pages     = {195-202},
  url       = {https://mlanthology.org/neurips/2005/cesabianchi2005neurips-improved/}
}