Abstract: Predictive models of human behavior--and in particular recommender systems--learn patterns from large volumes of historical activity data, in order to make personalized predictions that adapt to the needs, nuances, and preferences of individuals. Models may take incredibly complex data as input, ranging from text, images, social networks, or sequence data. However, the outputs they are trained to predict--clicks, purchases, transactions, etc.--are typically simple, numerical quantities, in order for the problem to be cast in terms of traditional supervised learning frameworks.

In this talk, we discuss possible extensions to such personalized, predictive models of human behavior so that they are capable of predicting complex structured outputs. For example, rather than training a model to predict what content a user might interact with, we could predict how they would react to unseen content, in the form of text they might write. Or, rather than predicting whether a user would purchase an existing product, we could predict the characteristics or attributes of the types of products that should be created.

bio: Julian McAuley has been an Assistant Professor in the Computer Science Department at the University of California, San Diego since 2014. Previously he was a postdoctoral scholar at Stanford University after receiving his PhD from the Australian National University in 2011. His research is concerned with developing predictive models of human behavior using large volumes of online activity data.