On Disentangled Representations Learned From Correlated Data
This addresses the gap in disentanglement research for real-world correlated data, which is incremental but important for applications like fairness.
The study analyzed how prominent disentanglement methods perform on correlated data, showing that induced correlations are learned and reflected in latent representations, with implications for fairness, and demonstrated resolution using weak supervision or post-hoc correction.
The focus of disentanglement approaches has been on identifying independent factors of variation in data. However, the causal variables underlying real-world observations are often not statistically independent. In this work, we bridge the gap to real-world scenarios by analyzing the behavior of the most prominent disentanglement approaches on correlated data in a large-scale empirical study (including 4260 models). We show and quantify that systematically induced correlations in the dataset are being learned and reflected in the latent representations, which has implications for downstream applications of disentanglement such as fairness. We also demonstrate how to resolve these latent correlations, either using weak supervision during training or by post-hoc correcting a pre-trained model with a small number of labels.