A Sober Look at the Unsupervised Learning of Disentangled Representations and their Evaluation

TL;DR

This paper critically examines the unsupervised learning of disentangled representations, showing theoretical limitations, extensive empirical results, and highlighting the importance of biases and supervision in achieving meaningful disentanglement.

Contribution

It provides a theoretical proof of the impossibility of unsupervised disentanglement without biases and offers a large-scale empirical study revealing limitations of current methods.

Findings

Disentanglement cannot be achieved without inductive biases.

Different evaluation metrics often disagree on disentanglement quality.

Higher disentanglement does not necessarily improve downstream task learning.

Abstract

The idea behind the \emph{unsupervised} learning of \emph{disentangled} representations is that real-world data is generated by a few explanatory factors of variation which can be recovered by unsupervised learning algorithms. In this paper, we provide a sober look at recent progress in the field and challenge some common assumptions. We first theoretically show that the unsupervised learning of disentangled representations is fundamentally impossible without inductive biases on both the models and the data. Then, we train over $14000$ models covering most prominent methods and evaluation metrics in a reproducible large-scale experimental study on eight data sets. We observe that while the different methods successfully enforce properties "encouraged" by the corresponding losses, well-disentangled models seemingly cannot be identified without supervision. Furthermore, different evaluation metrics do not always agree on what should be considered "disentangled" and exhibit systematic differences in the estimation. Finally, increased disentanglement does not seem to necessarily lead to a decreased sample complexity of learning for downstream tasks. Our results suggest that future work on disentanglement learning should be explicit about the role of inductive biases and (implicit) supervision, investigate concrete benefits of enforcing disentanglement of the learned representations, and consider a reproducible experimental setup covering several data sets.