Title: Vector-valued infinite task learning in style transfer
Authors: Zoltan Szabo - LSE (United Kingdom) [presenting]
Abstract: Style transfer is a central problem of machine learning. In various applications of style transfer, however, there is a continuum of styles to handle. We show how one can leverage vector-valued reproducing kernel Hilbert spaces and infinite task learning to tackle this challenge in a principled way. The approach is instantiated in emotion transfer, achieving low reconstruction cost on various benchmarks.