Deploying machine learning in production

Description

Predictive applications unlock tremendous value for businesses through intelligent use of data. From personalized recommenders, targeted advertising, to customer churn prediction and influence analysis, predictive applications can enhance the customer experience and increase business revenue. In order to make decisions such as which ad to show to which user, what items to recommend, and which customers are most likely to churn, these smart apps rely on sophisticated analysis machine learning algorithms and models. Machine learning techniques are powerful, but building and deploying such models require a lot of care and expertise.

With a new generation of tools, it is possible to build such applications with a minimal amount of algorithm expertise. However, one must know how to evaluate, test, and track the performance of these machine learning models over time. This talk engages potential application builders on topics such as common evaluation metrics, A/B testing set up, tracking model performance, tracking usage via real-time feedback, and updating models. The talk will go into details on how to read the tea leaves of model training and prediction output, and understand “distribution drift.” We will also show practical demonstrations of these ideas using GraphLab Create and other open-source tools.