Mapping the World of Music into Vector Spaces
Details
This month we have Ravi Mody and Tim Schmeier from iHeartRadio talk about their experiences building and applying models built on music data. The talk will take place at the iHeartRadio theater in Tribeca
Abstract:
In recent years the online music industry has used machine learning to transform the way people listen to and find new music. One of the most ubiquitous approaches has been serving recommendations to users using factorization models. These models have performed remarkably well because they capture complex user behavior in a much simpler vector space, enabling cheap but powerful predictions.
In this talk we'll discuss how we've generalized and extended this concept. We'll describe how we've been using collective matrix factorization and deep neural networks to map our user and audio data into a small number of vector spaces. We'll also discuss a simple API to interact with these vector spaces that covers a wide range of applications, motivated by real-world examples. We believe that this approach has far-reaching potential both inside and outside the music industry.
Bios:
Ravi Mody is Director of Data Science at iHeartRadio. He concentrates on using machine learning to elegantly solve real-world problems in production environments. He has worked as a scientist/engineer in a wide variety of industries, including financial services, digital advertising, and media. He has an M.S. in Computer Science and Machine Learning from UCSD, and B.S. in Computer Engineering from Johns Hopkins University where he conducted research on neural networks for computer vision.
Tim Schmeier is a Data Scientist at iHeartRadio where he works on a range of projects including music recommendations and user retention. Before coming to iHeartRadio he was an academic scientist conducting research in the fields of Neuroscience and Chemistry. Tim received a Ph.D. in Chemistry from Yale where he developed and optimized homogenous catalysts for the capture and conversion of carbon dioxide into useful chemical feedstocks.