Graphical Models for Characterizing Context

Опубликовано 17 августа 2016, 3:26

In this talk, I will cover a number of uses of constraints and context for machine learning tasks. Our core technical contribution is a unified approach that handles constraints and context to jointly tags examples across multiple domains. We will specifically focus on the task of face recognition, where besides low level appearance features, the information from multiple domains are captured using features derived from location, time and co-occurrence context. The heart of our approach is a generic probabilistic model of context that couples the domains through a set of cross-domain relations. The uncertainty estimate provided by the model naturally allows for active learning paradigms where the user is consulted after each iteration to tag additional faces. For more information see the 'Links' TAB. This is joint work with Simon Baker, Dahua Lin, Gang Hua and Amir Akbarzadeh.