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

Autoencoders and generative models produce some of the most spectacular deeplearning results to date. However, understanding and controlling the latentspace of these models presents a considerable challenge. Drawing inspirationfrom principal component analysis and autoencoder, we propose the PrincipalComponent Analysis Autoencoder (PCAAE). This is a novel autoencoder whoselatent space verifies two properties. Firstly, the dimensions are organised indecreasing importance with respect to the data at hand. Secondly, thecomponents of the latent space are statistically independent. We achieve thisby progressively increasing the latent space during training, and with acovariance loss applied to the latent codes. The resulting autoencoder producesa latent space which separates the intrinsic attributes of the data intodifferent components of the latent space, in a completely unsupervised manner.We also describe an extension of our approach to the case of powerful,pre-trained GANs. We show results on both synthetic examples of shapes and on astate-of-the-art GAN. For example, we are able to separate the color shadescale of hair and skin, pose of faces and the gender in the CelebA, withoutaccessing any labels. We compare the PCAAE with other state-of-the-artapproaches, in particular with respect to the ability to disentangle attributesin the latent space. We hope that this approach will contribute to betterunderstanding of the intrinsic latent spaces of powerful deep generativemodels.

PCAAE: Principal Component Analysis Autoencoder for organising the latent space of generative networks