What We Don't C: Manifold Disentanglement for Structured Discovery

TL;DR

This paper presents What We Don't C, a method that disentangles latent representations by removing conditioned information, enabling better exploration of unrepresented factors in high-dimensional data.

Contribution

It introduces a novel latent flow matching technique for explicit disentanglement of latent subspaces by removing conditional information.

Findings

Enhances interpretability of latent representations.

Facilitates discovery of unmodeled data factors.

Provides a simple mechanism for analyzing and controlling generative models.

Abstract

Accessing information in learned representations is critical for annotation, discovery, and data filtering in disciplines where high-dimensional datasets are common. We introduce What We Don't C, a novel approach based on latent flow matching that disentangles latent subspaces by explicitly removing information included in conditional guidance, resulting in meaningful residual representations. This allows factors of variation which have not already been captured in conditioning to become more readily available. We show how guidance in the flow path necessarily represses the information from the guiding, conditioning variables. Our results highlight this approach as a simple yet powerful mechanism for analyzing, controlling, and repurposing latent representations, providing a pathway toward using generative models to explore what we don't capture, consider, or catalog.