Creating a real-world machine learning product requires considerations beyond implementation of the machine learning algorithms itself. For example, very often the production environment is a constantly shifting data landscape. A well tested, carefully constructed model can become stale in a matter of days or even hours when data distribution drifts over time. In addition, a sustainable machine learning system needs to run on a healthy data ecosystem where bias is removed or accounted for as much as possible. Finally, evaluating, A/B testing, and launching machine learning product requires considerations very different from conventional product features.
In this TechTalk, we will share a few lessons learned from designing and maintaining a machine learning-minded product, and a product-minded machine learning system.
Grace Huang currently heads the Discovery data science team at Pinterest. Her team collaborates with engineering and product teams to create key machine learning products that power the search, recommendation, and visual discovery experience at Pinterest. In the past, Grace has led a wide range of data science projects spanning recommendation systems, search relevance, growth, and algorithm developments in genome sequencing as well as cancer diagnostics. Her passion lies in the space where algorithm meets real world application and product design. She holds a PhD in Computational Genomics from the Joint CMU-Pitt Program in Computational Biology.
