Improved Mixed-Example Data Augmentation

Show simple item record Summers, C en Dinneen, Michael en
dc.coverage.spatial Waikoloa Village, HI en 2019-03-27T21:09:21Z en 2019 en
dc.identifier.citation Proceedings - 2019 IEEE Winter Conference on Applications of Computer Vision, WACV 2019. IEEE. 9 pages. 2019 en
dc.identifier.isbn 978-1-7281-1975-5 en
dc.identifier.uri en
dc.description.abstract In order to reduce overfitting, neural networks are typically trained with data augmentation, the practice of artificially generating additional training data via label-preserving transformations of existing training examples. While these types of transformations make intuitive sense, recent work has demonstrated that even non-label-preserving data augmentation can be surprisingly effective, examining this type of data augmentation through linear combinations of pairs of examples. Despite their effectiveness, little is known about why such methods work. In this work, we aim to explore a new, more generalized form of this type of data augmentation in order to determine whether such linearity is necessary. By considering this broader scope of "mixed-example data augmentation", we find a much larger space of practical augmentation techniques, including methods that improve upon previous state-of-the-art. This generalization has benefits beyond the promise of improved performance, revealing a number of types of mixed-example data augmentation that are radically different from those considered in prior work, which provides evidence that current theories for the effectiveness of such methods are incomplete and suggests that any such theory must explain a much broader phenomenon. en
dc.description.uri en
dc.publisher IEEE en
dc.relation.ispartof 19th IEEE Winter Conference on Applications of Computer Vision (WACV) en
dc.relation.ispartofseries Proceedings - 2019 IEEE Winter Conference on Applications of Computer Vision, WACV 2019 en
dc.rights Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated. Previously published items are made available in accordance with the copyright policy of the publisher. en
dc.rights.uri en
dc.rights.uri en
dc.title Improved Mixed-Example Data Augmentation en
dc.type Conference Item en
dc.identifier.doi 10.1109/WACV.2019.00139 en
dc.rights.holder Copyright: IEEE en
pubs.finish-date 2019-01-11 en
pubs.start-date 2019-01-07 en
dc.rights.accessrights en
pubs.subtype Proceedings en
pubs.elements-id 744284 en Science en School of Computer Science en
pubs.arxiv-id 1805.11272 en
dc.identifier.eissn 2472-6737 en

Files in this item

Find Full text

This item appears in the following Collection(s)

Show simple item record


Search ResearchSpace