Coordinated Disentanglement with Iterative Mode Discovery Under Hidden Correlations

2026-07-19

AI Summary

This work addresses the challenge that conventional disentanglement methods often fail to preserve latent structural patterns when attributes exhibit hidden correlations. To overcome this limitation, the authors propose CoDID, an end-to-end framework that jointly performs iterative latent pattern discovery and enforces pattern-based conditional independence constraints. CoDID incorporates a dynamic neural architecture capable of adapting to varying numbers of latent patterns and introduces a meta-optimization–based coordination mechanism to mitigate error propagation across modules. Experimental results demonstrate that CoDID achieves state-of-the-art performance across multiple tasks, effectively balancing high-quality disentanglement with faithful retention of informative latent patterns.

Abstract

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.

Problem

• disentangled representation learning
• hidden correlations
• mode discovery
• attribute correlations
• conditional independence

Innovation

• disentangled representation learning
• hidden correlations
• iterative mode discovery
• meta-optimization
• conditional independence