Coordinated Disentanglement with Iterative Mode Discovery Under Hidden Correlations

Disentangled representation learning is a powerful paradigm for robust attribute prediction. While recent methods address attribute correlations, hidden correlations remain underexplored, where data under the value of a certain attribute exhibit underlying modes correlated with other attributes. To preserve mode information and achieve disentanglement, we jointly discover modes and enforce mode-based conditional independence. Yet, the interdependency between these two modules may lead to error amplification under naive iterations. We propose Coordinated Disentanglement with Iterative mode Discovery (CoDID), an end-to-end framework featuring a dynamic architecture that adapts to evolving number of modes, and a coordination mechanism that mitigates error amplification via meta-optimization. Empirical results demonstrate the state-of-the-art performance on diverse tasks.

Learning disentangled representations for correlated data has recently emerged as a challenging problem, where hidden correlations under the values of certain attributes remain underexplored. We propose an iterative framework that identifies underlying modes of certain attributes that exhibit hidden correlations, while learning the disentangled representations of these attributes that preserve mode information. This facilitates disentanglement for various downstream tasks involving high-level semantics and underlying correlations.