Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations

TL;DR

This paper critically examines the assumptions in unsupervised learning of disentangled representations, showing theoretical limitations and extensive empirical evidence that disentanglement often requires supervision and does not necessarily improve downstream task learning.

Contribution

It provides a theoretical proof of the impossibility of fully unsupervised disentanglement and presents a large-scale empirical study challenging common claims about disentanglement benefits.

Findings

Disentanglement cannot be achieved without inductive biases or supervision.

Different methods enforce properties but do not produce well-disentangled models without supervision.

Higher disentanglement does not reduce sample complexity for downstream tasks.

Abstract

The key idea behind the unsupervised learning of 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 more than 12000 models covering most prominent methods and evaluation metrics in a reproducible large-scale experimental study on seven different 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, increased disentanglement does not seem to 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.