Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness

TL;DR

This paper introduces a causal framework for evaluating disentangled representations in neural networks, addressing the lack of standardized validation methods and enabling robust, data-efficient model assessment.

Contribution

It proposes a causal perspective on representation learning, introduces a new metric for disentanglement and robustness, and provides an efficient estimation algorithm from observational data.

Findings

New causal metric for representation evaluation

Efficient linear-scaling estimation algorithm

Framework unifies disentanglement and domain shift robustness

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.