Machine learning engines are often designed to require little human input. However, while machines are fast, scalable, and reliable, selectively leveraging human input can be an excellent resource to provide continuous feedback and improvement. This is especially important given the inevitable changes in input data and the long tail of edge cases, which are traditionally difficult for machine learning engines. This talk discusses how we built a human-in-the-loop product categorization engine at Slice that involves different types of human workers - engineers, analysts, outsourced and crowdsourced workers - to allow for continuous improvement of the system with limited resources.
Conal Sathi is the Data Alchemist at Slice Technologies, where he is responsible for using machine learning and data mining algorithms to bring structure and deeper insights to consumer purchase data. At Slice, he was hired as the first machine learning engineer, where he built end-to-end systems from R&D to production, involving natural language processing, graph mining, and crowdsourcing. Now he leads a team of stellar machine learning engineers to build the world's largest purchase graph. Prior to Slice, Conal focused on research in the intersection of social network analysis and natural language processing. He examined how sentiment flows through hyperlink networks and explored how to create a content prediction system for Twitter users. Conal has a M.S. in Computer Science with a focus in Artificial Intelligence and a B.S. in Symbolic Systems with a focus in Neuroscience from Stanford University.