IPAB Workshop - 03/05/2018

Talk Title: Meta Learning for Domain Generalisation

Talk Abstract: Machine learning conventionally assumes that training and testing data are drawn from the same distribution. However it is not always possible to guarantee this, leading to the problem of domain shift that occurs when this assumption is violated. Domain Generalization (DG) techniques attempt to alleviate this issue by building models which by design generalize well to novel testing domains that differ from the training set. In this talk, Tim will introduce a few popular DG methods, and recent work on meta-learning for domain-generalisation with applications both in computer vision and reinforcement learning for control.