DIVA: Domain Invariant Variational Autoencoders

Maximilian Ilse, Jakub M. Tomczak, Christos Louizos, Max Welling
; Proceedings of the Third Conference on Medical Imaging with Deep Learning, PMLR 121:322-348, 2020.

Abstract

We consider the problem of domain generalization, namely, how to learn representations given data from a set of domains that generalize to data from a previously unseen domain. We propose the Domain Invariant Variational Autoencoder (DIVA), a generative model that tackles this problem by learning three independent latent subspaces, one for the domain, one for the class, and one for any residual variations. We highlight that due to the generative nature of our model we can also incorporate unlabeled data from known or previously unseen domains. To the best of our knowledge this has not been done before in a domain generalization setting. This property is highly desirable in fields like medical imaging where labeled data is scarce. We experimentally evaluate our model on the rotated MNIST benchmark and a malaria cell images dataset where we show that (i) the learned subspaces are indeed complementary to each other, (ii) we improve upon recent works on this task and (iii) incorporating unlabelled data can boost the performance even further.

Cite this Paper


BibTeX
@InProceedings{pmlr-v121-ilse20a, title = {DIVA: Domain Invariant Variational Autoencoders}, author = {Ilse, Maximilian and Tomczak, Jakub M. and Louizos, Christos and Welling, Max}, pages = {322--348}, year = {2020}, editor = {Tal Arbel and Ismail Ben Ayed and Marleen de Bruijne and Maxime Descoteaux and Herve Lombaert and Christopher Pal}, volume = {121}, series = {Proceedings of Machine Learning Research}, address = {Montreal, QC, Canada}, month = {06--08 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v121/ilse20a/ilse20a.pdf}, url = {http://proceedings.mlr.press/v121/ilse20a.html}, abstract = {We consider the problem of domain generalization, namely, how to learn representations given data from a set of domains that generalize to data from a previously unseen domain. We propose the Domain Invariant Variational Autoencoder (DIVA), a generative model that tackles this problem by learning three independent latent subspaces, one for the domain, one for the class, and one for any residual variations. We highlight that due to the generative nature of our model we can also incorporate unlabeled data from known or previously unseen domains. To the best of our knowledge this has not been done before in a domain generalization setting. This property is highly desirable in fields like medical imaging where labeled data is scarce. We experimentally evaluate our model on the rotated MNIST benchmark and a malaria cell images dataset where we show that (i) the learned subspaces are indeed complementary to each other, (ii) we improve upon recent works on this task and (iii) incorporating unlabelled data can boost the performance even further.} }
Endnote
%0 Conference Paper %T DIVA: Domain Invariant Variational Autoencoders %A Maximilian Ilse %A Jakub M. Tomczak %A Christos Louizos %A Max Welling %B Proceedings of the Third Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2020 %E Tal Arbel %E Ismail Ben Ayed %E Marleen de Bruijne %E Maxime Descoteaux %E Herve Lombaert %E Christopher Pal %F pmlr-v121-ilse20a %I PMLR %J Proceedings of Machine Learning Research %P 322--348 %U http://proceedings.mlr.press %V 121 %W PMLR %X We consider the problem of domain generalization, namely, how to learn representations given data from a set of domains that generalize to data from a previously unseen domain. We propose the Domain Invariant Variational Autoencoder (DIVA), a generative model that tackles this problem by learning three independent latent subspaces, one for the domain, one for the class, and one for any residual variations. We highlight that due to the generative nature of our model we can also incorporate unlabeled data from known or previously unseen domains. To the best of our knowledge this has not been done before in a domain generalization setting. This property is highly desirable in fields like medical imaging where labeled data is scarce. We experimentally evaluate our model on the rotated MNIST benchmark and a malaria cell images dataset where we show that (i) the learned subspaces are indeed complementary to each other, (ii) we improve upon recent works on this task and (iii) incorporating unlabelled data can boost the performance even further.
APA
Ilse, M., Tomczak, J.M., Louizos, C. & Welling, M.. (2020). DIVA: Domain Invariant Variational Autoencoders. Proceedings of the Third Conference on Medical Imaging with Deep Learning, in PMLR 121:322-348

Related Material