by Maximilian Ilse (ilse.maximilian@gmail.com) and Jakub M. Tomczak, Christos Louizos and Max Welling

As described in Appendix 5.1.7 while losing a good amount of interpretability, DIVA works even without z_d. Resulting in a, especially from an optimization perspective, simpler model. The code for a simpler DIVA is included in this repository as well.

PyTorch implementation of our paper 'DIVA: Domain Invariant Variational Autoencoders'

• Ilse, M., Tomczak, J. M., C. Louizos & Welling, M. (2018). DIVA: Domain Invariant Variational Autoencoders. arXiv preprint arXiv:1905.10427. link.

• Python 3.6
• PyTorch 1.0.1
• Scikit-image 0.14.1
• Scikit-learn 0.19.1
• Seaborn 0.9.0

The malaria cell images dataset can be found here: https://lhncbc.nlm.nih.gov/publication/pub9932

If you find any bugs or have any questions about this code please contact Maximilian. We cannot guarantee any support for this software.

Please cite our paper if you use this code in your research:

@article{ilse_diva:_2019,
	title = {{DIVA}: {Domain} {Invariant} {Variational} {Autoencoders}},
	author = {Ilse, Maximilian and Tomczak, Jakub M. and Louizos, Christos and Welling, Max},
	journal = {arXiv:1905.10427 [cs, stat]},
	year = {2019}
}

The work conducted by Maximilian Ilse was funded by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Grant DLMedIa: Deep Learning for Medical Image Analysis).

The work conducted by Jakub Tomczak was funded by the European Commission within the Marie Skodowska-Curie Individual Fellowship (Grant No. 702666, ”Deep learning and Bayesian inference for medical imaging”).