Recommendation Systems in Real Life

Monday, March 25, 2019 - 1:25pm - 2:25pm
Lind 305
Mark Hsiao (Netflix, Inc)
Recommendation systems are used everywhere nowadays. This talk will discuss how differently recommendation systems are researched and applied in academia and industry, and challenges one might face or criteria one might care about when building recommendation systems. In real life, the job of a recommendation algorithm is not to be very predictive of past held-out interactions but is to be predictive of future interactions. What is really important is its performance after being deployed, which is in the future. In addition, users change over time, items change over time, even the system around the recommendation algorithm is going to change over time, which might impact things and need to be considered. In this talk, we will talk about how we tackle some of these challenges at Netflix.

Ko-Jen (Mark) Hsiao is a senior research scientist at Netflix, where he researches and implements innovative algorithms to optimize Netflix's core personalization. He has extensive experiences in applying machine learning at scale, building and A/B testing ranking and recommendation systems. Before he joined Netflix, he was a data scientist at Whisper, an anonymous social network app with over 30 million monthly active users. He developed and implemented several end-to-end machine learning systems used by the company. He obtained his PhD from the University of Michigan, where he focused on combining disparate information for machine learning applications.