Disentangling shared and private latent factors in multimodal Variational Autoencoders

Kaspar Märtens, Christopher Yau
Proceedings of the 18th Machine Learning in Computational Biology meeting, PMLR 240:60-75, 2024.

Abstract

Generative models for multimodal data permit the identification of latent factors that may be associated with important determinants of observed data heterogeneity. Common or shared factors could be important for explaining variation across modalities whereas other factors may be private and important only for the explanation of a single modality. Multimodal Variational Autoencoders, such as MVAE and MMVAE, are a natural choice for inferring those underlying latent factors and separating shared variation from private. In this work, we investigate their capability to reliably perform this disentanglement. In particular, we highlight a challenging problem setting where modality-specific variation dominates the shared signal. Taking a cross-modal prediction perspective, we demonstrate limitations of existing models, and propose a modification how to make them more robust to modality-specific variation. Our findings are supported by experiments on synthetic as well as various real-world multi-omics data sets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v240-martens24a, title = {Disentangling shared and private latent factors in multimodal Variational Autoencoders}, author = {M\"artens, Kaspar and Yau, Christopher}, booktitle = {Proceedings of the 18th Machine Learning in Computational Biology meeting}, pages = {60--75}, year = {2024}, editor = {Knowles, David A. and Mostafavi, Sara}, volume = {240}, series = {Proceedings of Machine Learning Research}, month = {30 Nov--01 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v240/martens24a/martens24a.pdf}, url = {https://proceedings.mlr.press/v240/martens24a.html}, abstract = {Generative models for multimodal data permit the identification of latent factors that may be associated with important determinants of observed data heterogeneity. Common or shared factors could be important for explaining variation across modalities whereas other factors may be private and important only for the explanation of a single modality. Multimodal Variational Autoencoders, such as MVAE and MMVAE, are a natural choice for inferring those underlying latent factors and separating shared variation from private. In this work, we investigate their capability to reliably perform this disentanglement. In particular, we highlight a challenging problem setting where modality-specific variation dominates the shared signal. Taking a cross-modal prediction perspective, we demonstrate limitations of existing models, and propose a modification how to make them more robust to modality-specific variation. Our findings are supported by experiments on synthetic as well as various real-world multi-omics data sets. } }
Endnote
%0 Conference Paper %T Disentangling shared and private latent factors in multimodal Variational Autoencoders %A Kaspar Märtens %A Christopher Yau %B Proceedings of the 18th Machine Learning in Computational Biology meeting %C Proceedings of Machine Learning Research %D 2024 %E David A. Knowles %E Sara Mostafavi %F pmlr-v240-martens24a %I PMLR %P 60--75 %U https://proceedings.mlr.press/v240/martens24a.html %V 240 %X Generative models for multimodal data permit the identification of latent factors that may be associated with important determinants of observed data heterogeneity. Common or shared factors could be important for explaining variation across modalities whereas other factors may be private and important only for the explanation of a single modality. Multimodal Variational Autoencoders, such as MVAE and MMVAE, are a natural choice for inferring those underlying latent factors and separating shared variation from private. In this work, we investigate their capability to reliably perform this disentanglement. In particular, we highlight a challenging problem setting where modality-specific variation dominates the shared signal. Taking a cross-modal prediction perspective, we demonstrate limitations of existing models, and propose a modification how to make them more robust to modality-specific variation. Our findings are supported by experiments on synthetic as well as various real-world multi-omics data sets.
APA
Märtens, K. & Yau, C.. (2024). Disentangling shared and private latent factors in multimodal Variational Autoencoders. Proceedings of the 18th Machine Learning in Computational Biology meeting, in Proceedings of Machine Learning Research 240:60-75 Available from https://proceedings.mlr.press/v240/martens24a.html.

Related Material