Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness

Raphael Suter, Djordje Miladinovic, Bernhard Schölkopf, Stefan Bauer
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:6056-6065, 2019.

Abstract

The ability to learn disentangled representations that split underlying sources of variation in high dimensional, unstructured data is important for data efficient and robust use of neural networks. While various approaches aiming towards this goal have been proposed in recent times, a commonly accepted definition and validation procedure is missing. We provide a causal perspective on representation learning which covers disentanglement and domain shift robustness as special cases. Our causal framework allows us to introduce a new metric for the quantitative evaluation of deep latent variable models. We show how this metric can be estimated from labeled observational data and further provide an efficient estimation algorithm that scales linearly in the dataset size.

Cite this Paper


BibTeX
@InProceedings{pmlr-v97-suter19a, title = {Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness}, author = {Suter, Raphael and Miladinovic, Djordje and Sch{\"o}lkopf, Bernhard and Bauer, Stefan}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {6056--6065}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/suter19a/suter19a.pdf}, url = {https://proceedings.mlr.press/v97/suter19a.html}, abstract = {The ability to learn disentangled representations that split underlying sources of variation in high dimensional, unstructured data is important for data efficient and robust use of neural networks. While various approaches aiming towards this goal have been proposed in recent times, a commonly accepted definition and validation procedure is missing. We provide a causal perspective on representation learning which covers disentanglement and domain shift robustness as special cases. Our causal framework allows us to introduce a new metric for the quantitative evaluation of deep latent variable models. We show how this metric can be estimated from labeled observational data and further provide an efficient estimation algorithm that scales linearly in the dataset size.} }
Endnote
%0 Conference Paper %T Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness %A Raphael Suter %A Djordje Miladinovic %A Bernhard Schölkopf %A Stefan Bauer %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-suter19a %I PMLR %P 6056--6065 %U https://proceedings.mlr.press/v97/suter19a.html %V 97 %X The ability to learn disentangled representations that split underlying sources of variation in high dimensional, unstructured data is important for data efficient and robust use of neural networks. While various approaches aiming towards this goal have been proposed in recent times, a commonly accepted definition and validation procedure is missing. We provide a causal perspective on representation learning which covers disentanglement and domain shift robustness as special cases. Our causal framework allows us to introduce a new metric for the quantitative evaluation of deep latent variable models. We show how this metric can be estimated from labeled observational data and further provide an efficient estimation algorithm that scales linearly in the dataset size.
APA
Suter, R., Miladinovic, D., Schölkopf, B. & Bauer, S.. (2019). Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:6056-6065 Available from https://proceedings.mlr.press/v97/suter19a.html.

Related Material