ASSET Seminar: Equivariance in Deep Learning, Kostas Daniilidis (University of Pennsylvania)
ASSET Seminar: Equivariance in Deep Learning, Kostas Daniilidis (University of Pennsylvania)
ABSTRACT Traditional convolutional networks exhibit unprecedented robustness to intraclass nuisances when trained on big data. Generalization with respect to geometric transformations has been achieved via expensive data augmentation. It has been shown recently that data augmentation can be avoided if networks are structured such that feature representations are transformed the same way as the input, a desirable property called equivariance. In […]