Disentangled Representation Learning via Modular Compositional Bias

TL;DR

This paper introduces a modular compositional bias approach for disentangled representation learning that uses factor-specific remixing strategies, enabling flexible and joint disentanglement of attributes and objects without changing architectures or objectives.

Contribution

It proposes a novel modular inductive bias framework that decouples from objectives and architectures, allowing flexible disentanglement through remixing strategies based on data recombination rules.

Findings

Achieves competitive attribute and object disentanglement.

Successfully disentangles global style and objects jointly.

Demonstrates flexibility across different factors of variation.

Abstract

Recent disentangled representation learning (DRL) methods heavily rely on factor specific strategies-either learning objectives for attributes or model architectures for objects-to embed inductive biases. Such divergent approaches result in significant overhead when novel factors of variation do not align with prior assumptions, such as statistical independence or spatial exclusivity, or when multiple factors coexist, as practitioners must redesign architectures or objectives. To address this, we propose a compositional bias, a modular inductive bias decoupled from both objectives and architectures. Our key insight is that different factors obey distinct recombination rules in the data distribution: global attributes are mutually exclusive, e.g., a face has one nose, while objects share a common support (any subset of objects can co-exist). We therefore randomly remix latents according to factor-specific rules, i.e., a mixing strategy, and force the encoder to discover whichever factor structure the mixing strategy reflects through two complementary objectives: (i) a prior loss that ensures every remix decodes into a realistic image, and (ii) the compositional consistency loss introduced by Wiedemer et al. (arXiv:2310.05327), which aligns each composite image with its corresponding composite latent. Under this general framework, simply adjusting the mixing strategy enables disentanglement of attributes, objects, and even both, without modifying the objectives or architectures. Extensive experiments demonstrate that our method shows competitive performance in both attribute and object disentanglement, and uniquely achieves joint disentanglement of global style and objects. Code is available at https://github.com/whieya/Compositional-DRL.