A Second-order Look at Stability and Generalization

Andreas Maurer
Proceedings of the 2017 Conference on Learning Theory, PMLR 65:1461-1475, 2017.

Abstract

Using differentiability assumptions on the loss function and a concentration inequality for bounded second order differences it is shown that the generalization error for classification with L2 regularisation obeys a Bernstein-type inequality.

Cite this Paper


BibTeX
@InProceedings{pmlr-v65-maurer17a, title = {A Second-order Look at Stability and Generalization}, author = {Maurer, Andreas}, booktitle = {Proceedings of the 2017 Conference on Learning Theory}, pages = {1461--1475}, year = {2017}, editor = {Kale, Satyen and Shamir, Ohad}, volume = {65}, series = {Proceedings of Machine Learning Research}, month = {07--10 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v65/maurer17a/maurer17a.pdf}, url = { http://proceedings.mlr.press/v65/maurer17a.html }, abstract = {Using differentiability assumptions on the loss function and a concentration inequality for bounded second order differences it is shown that the generalization error for classification with L2 regularisation obeys a Bernstein-type inequality.} }
Endnote
%0 Conference Paper %T A Second-order Look at Stability and Generalization %A Andreas Maurer %B Proceedings of the 2017 Conference on Learning Theory %C Proceedings of Machine Learning Research %D 2017 %E Satyen Kale %E Ohad Shamir %F pmlr-v65-maurer17a %I PMLR %P 1461--1475 %U http://proceedings.mlr.press/v65/maurer17a.html %V 65 %X Using differentiability assumptions on the loss function and a concentration inequality for bounded second order differences it is shown that the generalization error for classification with L2 regularisation obeys a Bernstein-type inequality.
APA
Maurer, A.. (2017). A Second-order Look at Stability and Generalization. Proceedings of the 2017 Conference on Learning Theory, in Proceedings of Machine Learning Research 65:1461-1475 Available from http://proceedings.mlr.press/v65/maurer17a.html .

Related Material