English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this paper, we present an in-depth investigation of the convolutionalautoencoder (CAE) bottleneck. Autoencoders (AE), and especially theirconvolutional variants, play a vital role in the current deep learning toolbox.Researchers and practitioners employ CAEs for a variety of tasks, ranging fromoutlier detection and compression to transfer and representation learning.Despite their widespread adoption, we have limited insight into how thebottleneck shape impacts the emergent properties of the CAE. We demonstratethat increased height and width of the bottleneck drastically improvesgeneralization, which in turn leads to better performance of the latent codesin downstream transfer learning tasks. The number of channels in thebottleneck, on the other hand, is secondary in importance. Furthermore, we showempirically that, contrary to popular belief, CAEs do not learn to copy theirinput, even when the bottleneck has the same number of neurons as there arepixels in the input. Copying does not occur, despite training the CAE for 1,000epochs on a tiny ($\approx$ 600 images) dataset. We believe that the findingsin this paper are directly applicable and will lead to improvements in modelsthat rely on CAEs.

Walking the Tightrope: An Investigation of the Convolutional Autoencoder Bottleneck