Poster
Safe Grid Search with Optimal Complexity
Eugene Ndiaye · Tam Le · Olivier Fercoq · Joseph Salmon · Ichiro Takeuchi

Tue Jun 11th 06:30 -- 09:00 PM @ Pacific Ballroom #None

Popular machine learning estimators involve regularization parameters that can be challenging to tune, and standard strategies rely on grid search for this task. In this paper, we revisit the techniques of approximating the regularization path up to predefined tolerance $\epsilon$ in a unified framework and show that its complexity is $O(1/\sqrt[d]{\epsilon})$ for uniformly convex loss of order $d \geq 2$ and $O(1/\sqrt{\epsilon})$ for Generalized Self-Concordant functions. This framework encompasses least-squares but also logistic regression, a case that as far as we know was not handled as precisely in previous works. We leverage our technique to provide refined bounds on the validation error as well as a practical algorithm for hyperparameter tuning. The latter has global convergence guarantee when targeting a prescribed accuracy on the validation set. Last but not least, our approach helps relieving the practitioner from the (often neglected) task of selecting a stopping criterion when optimizing over the training set: our method automatically calibrates this criterion based on the targeted accuracy on the validation set.

Author Information

Tam Le (RIKEN AIP)

My name is Tam Le. I officially received my PhD degree from Kyoto University in 01/2016, under the supervision of Marco Cuturi and Akihiro Yamamoto. Currently, I have been working as a postdoc researcher at RIKEN AIP, Japan, mentored by Makoto Yamada from September, 2017. Prior to this, I spent 1.5 year as a postdoc researcher at Nagoya Institute of Technology and National Institute of Materials Science, working with Ichiro Takeuchi.