Latent Space Factorisation and Manipulation via Matrix Subspace Projection

Xiao Li, Chenghua Lin, Ruizhe Li, Chaozheng Wang, Frank Guerin
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:5916-5926, 2020.

Abstract

We tackle the problem disentangling the latent space of an autoencoder in order to separate labelled attribute information from other characteristic information. This then allows us to change selected attributes while preserving other information. Our method, matrix subspace projection, is much simpler than previous approaches to latent space factorisation, for example not requiring multiple discriminators or a careful weighting among their loss functions. Furthermore our new model can be applied to autoencoders as a plugin, and works across diverse domains such as images or text. We demonstrate the utility of our method for attribute manipulation in autoencoders trained across varied domains, using both human evaluation and automated methods. The quality of generation of our new model (e.g. reconstruction, conditional generation) is highly competitive to a number of strong baselines.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-li20i, title = {Latent Space Factorisation and Manipulation via Matrix Subspace Projection}, author = {Li, Xiao and Lin, Chenghua and Li, Ruizhe and Wang, Chaozheng and Guerin, Frank}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {5916--5926}, year = {2020}, editor = {Hal Daumé III and Aarti Singh}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/li20i/li20i.pdf}, url = { http://proceedings.mlr.press/v119/li20i.html }, abstract = {We tackle the problem disentangling the latent space of an autoencoder in order to separate labelled attribute information from other characteristic information. This then allows us to change selected attributes while preserving other information. Our method, matrix subspace projection, is much simpler than previous approaches to latent space factorisation, for example not requiring multiple discriminators or a careful weighting among their loss functions. Furthermore our new model can be applied to autoencoders as a plugin, and works across diverse domains such as images or text. We demonstrate the utility of our method for attribute manipulation in autoencoders trained across varied domains, using both human evaluation and automated methods. The quality of generation of our new model (e.g. reconstruction, conditional generation) is highly competitive to a number of strong baselines.} }
Endnote
%0 Conference Paper %T Latent Space Factorisation and Manipulation via Matrix Subspace Projection %A Xiao Li %A Chenghua Lin %A Ruizhe Li %A Chaozheng Wang %A Frank Guerin %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-li20i %I PMLR %P 5916--5926 %U http://proceedings.mlr.press/v119/li20i.html %V 119 %X We tackle the problem disentangling the latent space of an autoencoder in order to separate labelled attribute information from other characteristic information. This then allows us to change selected attributes while preserving other information. Our method, matrix subspace projection, is much simpler than previous approaches to latent space factorisation, for example not requiring multiple discriminators or a careful weighting among their loss functions. Furthermore our new model can be applied to autoencoders as a plugin, and works across diverse domains such as images or text. We demonstrate the utility of our method for attribute manipulation in autoencoders trained across varied domains, using both human evaluation and automated methods. The quality of generation of our new model (e.g. reconstruction, conditional generation) is highly competitive to a number of strong baselines.
APA
Li, X., Lin, C., Li, R., Wang, C. & Guerin, F.. (2020). Latent Space Factorisation and Manipulation via Matrix Subspace Projection. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:5916-5926 Available from http://proceedings.mlr.press/v119/li20i.html .

Related Material