From attribute to zero-shot learning
Speaker: Timothy Hospedales, assistant professor at Queen Mary University, London, U.K
Abstract:The learning of attributes, and the related task of zero-shot learning are topical areas in computer vision and beyond. In this talk I will introduce these areas for a general audience before presenting some specific recent progress made by our group. This will include scaling up with weak supervision, transfer learning for dealing with domain shift and connecting multiple modalities with multi-view learning. Along the way I will demonstrate applications in a variety of areas including fine-grained image retrieval, person re-identification, action recognition and mapping text to attribute ontologies.