Show simple item record

dc.contributor.authorGarriga-Alonso, A
dc.contributor.authorAitchison, L
dc.contributor.authorRasmussen, CE
dc.date.accessioned2019-08-05T23:30:05Z
dc.date.available2019-08-05T23:30:05Z
dc.date.issued2018-08-16
dc.identifier.urihttps://www.repository.cam.ac.uk/handle/1810/295286
dc.description.abstractWe show that the output of a (residual) convolutional neural network (CNN) with an appropriate prior over the weights and biases is a Gaussian process (GP) in the limit of infinitely many convolutional filters, extending similar results for dense networks. For a CNN, the equivalent kernel can be computed exactly and, unlike "deep kernels", has very few parameters: only the hyperparameters of the original CNN. Further, we show that this kernel has two properties that allow it to be computed efficiently; the cost of evaluating the kernel for a pair of images is similar to a single forward pass through the original CNN with only one filter per layer. The kernel equivalent to a 32-layer ResNet obtains 0.84% classification error on MNIST, a new record for GPs with a comparable number of parameters.
dc.rightsAll rights reserved
dc.titleDeep convolutional networks as shallow Gaussian processes
dc.typeArticle
prism.publicationDate2019
prism.publicationName7th International Conference on Learning Representations, ICLR 2019
dc.identifier.doi10.17863/CAM.42340
dcterms.dateAccepted2018-08-16
rioxxterms.version
rioxxterms.licenseref.urihttp://www.rioxx.net/licenses/all-rights-reserved
rioxxterms.licenseref.startdate2019-01-01
dc.contributor.orcidRasmussen, Carl [0000-0001-8899-7850]
rioxxterms.typeJournal Article/Review
cam.issuedOnline2018-08-16
cam.orpheus.success2022-03-03: removed from embargo task
cam.orpheus.counter94


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record