Traditionally, pinpointing the best possible model for a given prediction problem requires optimizing an entire parameter space—and even taking this step involves choosing between a number of different methodologies. Auto Tune Models (ATM), our cloud-based modeling system, provides an easier, faster way to narrow down to the best choice. ATM performs bandit-based and Gaussian process learning to decide among methodologies, as well as to pinpoint which parameters and hyperparameters should be used for modeling. The result is a vastly more efficient end-to-end process.

github

hdi

Publications

ATM: A distributed, collaborative, scalable system for automated machine learning (PDF)

Thomas Swearingen, Will Drevo, Bennett Cyphers, Alfredo Cuesta-Infante, Kalyan Veeramachaneni. IEEE International Conference on Big Data, Boston, 2017.

Comparing AutoML solutions to human baselines (PDF)

Thomas Swearingen, Bennett Cyphers, Arun Ross, Gaurav Sheni, Kalyan Veeramachaneni

Contributors

Bennett Cyphers
Thomas Swearingen
Alfredo Cuesta-Infante
Will Drevo

FeatureHub

SenseML