Learning Optimal Linear Regularizers

Matthew Streeter
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:5996-6004, 2019.

Abstract

We present algorithms for efficiently learning regularizers that improve generalization. Our approach is based on the insight that regularizers can be viewed as upper bounds on the generalization gap, and that reducing the slack in the bound can improve performance on test data. For a broad class of regularizers, the hyperparameters that give the best upper bound can be computed using linear programming. Under certain Bayesian assumptions, solving the LP lets us "jump" to the optimal hyperparameters given very limited data. This suggests a natural algorithm for tuning regularization hyperparameters, which we show to be effective on both real and synthetic data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-streeter19a, title = {Learning Optimal Linear Regularizers}, author = {Streeter, Matthew}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {5996--6004}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/streeter19a/streeter19a.pdf}, url = {https://proceedings.mlr.press/v97/streeter19a.html}, abstract = {We present algorithms for efficiently learning regularizers that improve generalization. Our approach is based on the insight that regularizers can be viewed as upper bounds on the generalization gap, and that reducing the slack in the bound can improve performance on test data. For a broad class of regularizers, the hyperparameters that give the best upper bound can be computed using linear programming. Under certain Bayesian assumptions, solving the LP lets us "jump" to the optimal hyperparameters given very limited data. This suggests a natural algorithm for tuning regularization hyperparameters, which we show to be effective on both real and synthetic data.} }
Endnote
%0 Conference Paper %T Learning Optimal Linear Regularizers %A Matthew Streeter %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-streeter19a %I PMLR %P 5996--6004 %U https://proceedings.mlr.press/v97/streeter19a.html %V 97 %X We present algorithms for efficiently learning regularizers that improve generalization. Our approach is based on the insight that regularizers can be viewed as upper bounds on the generalization gap, and that reducing the slack in the bound can improve performance on test data. For a broad class of regularizers, the hyperparameters that give the best upper bound can be computed using linear programming. Under certain Bayesian assumptions, solving the LP lets us "jump" to the optimal hyperparameters given very limited data. This suggests a natural algorithm for tuning regularization hyperparameters, which we show to be effective on both real and synthetic data.
APA
Streeter, M.. (2019). Learning Optimal Linear Regularizers. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:5996-6004 Available from https://proceedings.mlr.press/v97/streeter19a.html.

Related Material