Seminar Series Archive
Julian McAuley
University of California, San Diego
November 22, 2019
11:00am - 12:00pm
Title:
Personalized Models of Natural Language
Abstract:
Abstract: Recommender systems seek to capture complex relationships between people and content they interact with. Other than their applications in traditional settings (like e-commerce), similar techniques are potentially impactful in a variety of domains where variance among individuals accounts for much of the variability in the data. Our lab has studied examples including personalized preference models in image and audio domains, personalized dialog systems, and personalized healthcare.
In this talk I'll discuss one such line of work that seeks to combine ideas from recommender systems with techniques from natural language processing. Examples include models for question answering, text and recipe generation, and interpretability. By personalizing these models, we are better able to adapt to both the preferences and the linguistic nuances of individuals. Doing so leads to language models that are higher fidelity both quantitatively and qualitatively, in terms of traditional and subjective metrics.
In this talk I'll discuss one such line of work that seeks to combine ideas from recommender systems with techniques from natural language processing. Examples include models for question answering, text and recipe generation, and interpretability. By personalizing these models, we are better able to adapt to both the preferences and the linguistic nuances of individuals. Doing so leads to language models that are higher fidelity both quantitatively and qualitatively, in terms of traditional and subjective metrics.
Speaker Bio:
Julian McAuley is an Associate Professor at UCSD. He has not won any scholarly awards and is not a member of any professional societies. He likes bicycling and baroque keyboard.