Show simple item record

dc.contributor.authorMouton, Coenraad
dc.contributor.authorMyburgh, Johannes C.
dc.contributor.authorDavel, Marelie H.
dc.date.accessioned2021-04-14T07:59:10Z
dc.date.available2021-04-14T07:59:10Z
dc.date.issued2020
dc.identifier.isbn978-3-030-66151-9
dc.identifier.issn1865-0929
dc.identifier.urihttp://hdl.handle.net/10394/36934
dc.identifier.urihttps://doi.org/10.1007/978-3-030-66151-9_17
dc.identifier.urihttps://arxiv.org/abs/2103.10097
dc.description.abstractConvolutional Neural Networks have become the standard for image classification tasks, however, these architectures are not invariant to translations of the input image. This lack of invariance is attributed to the use of stride which subsamples the input, resulting in a loss of information, and fully connected layers which lack spatial reasoning. We show that stride can greatly benefit translation invariance given that it is combined with sufficient similarity between neighbouring pixels, a characteristic which we refer to as local homogeneity. We also observe that this characteristic is dataset-specific and dictates the relationship between pooling kernel size and stride required for translation invariance. Furthermore we find that a trade-off exists between generalization and translation invariance in the case of pooling kernel size, as larger kernel sizes lead to better invariance but poorer generalization. Finally we explore the efficacy of other solutions proposed, namely global average pooling, anti-aliasing, and data augmentation, both empirically and through the lens of local homogeneity.en_US
dc.language.isoenen_US
dc.publisherSouthern African Conference for Artificial Intelligence Researchen_US
dc.subjectTranslation invarianceen_US
dc.subjectSubsamplingen_US
dc.subjectConvolutional Neural Networken_US
dc.subjectLocal homogeneityen_US
dc.titleStride and translation invariance in CNNsen_US
dc.typeArticleen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record