# Disentangling Granularity: An Implicit Inductive Bias in Factorized VAEs

arXiv: https://arxiv.org/abs/2505.24684 · HTML: https://mlsanity.com/paper/2505.24684/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 36 references that are in the map.
- First posted on arXiv 30 May 2025.

## Builds on (most cited first, 19 of 19)

- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations](https://mlsanity.com/paper/1811.12359/) (2018)
- [Disentangling by Factorising](https://mlsanity.com/paper/1802.05983/) (2018)
- [Variational Inference of Disentangled Latent Concepts from Unlabeled Observations](https://mlsanity.com/paper/1711.00848/) (2017)
- [What the DAAM: Interpreting Stable Diffusion Using Cross Attention](https://mlsanity.com/paper/2210.04885/) (2022)
- [Disentangled Representation Learning](https://mlsanity.com/paper/2211.11695/) (2022)
- [Masked Scene Contrast: A Scalable Framework for Unsupervised 3D Representation Learning](https://mlsanity.com/paper/2303.14191/) (2023)
- [A Sober Look at the Unsupervised Learning of Disentangled Representations and their Evaluation](https://mlsanity.com/paper/2010.14766/) (2020)
- [Micron-BERT: BERT-based Facial Micro-Expression Recognition](https://mlsanity.com/paper/2304.03195/) (2023)
- [Learning Disentangled Representation by Exploiting Pretrained Generative Models: A Contrastive Learning View](https://mlsanity.com/paper/2102.10543/) (2021)
- [Learning Background Prompts to Discover Implicit Knowledge for Open Vocabulary Object Detection](https://mlsanity.com/paper/2406.00510/) (2024)
- [Neural Speech Phase Prediction based on Parallel Estimation Architecture and Anti-Wrapping Losses](https://mlsanity.com/paper/2211.15974/) (2022)
- [Beyond Alignment: Blind Video Face Restoration via Parsing-Guided Temporal-Coherent Transformer](https://mlsanity.com/paper/2404.13640/) (2024)
- [Multi-View Attentive Contextualization for Multi-View 3D Object Detection](https://mlsanity.com/paper/2405.12200/) (2024)
- [Recent Advances in Deep Learning Techniques for Face Recognition](https://mlsanity.com/paper/2103.10492/) (2021)
- [Robust Ordinal VAE: Employing Noisy Pairwise Comparisons for Disentanglement](https://mlsanity.com/paper/1910.05898/) (2019)
- [On the Bias Against Inductive Biases](https://mlsanity.com/paper/2105.14077/) (2021)
- [Speech Reconstruction from Silent Tongue and Lip Articulation By Pseudo Target Generation and Domain Adversarial Training](https://mlsanity.com/paper/2304.05574/) (2023)
- [Disentangling Latent Factors of Variational Auto-Encoder with Whitening](https://mlsanity.com/paper/1811.03444/) (2018)

## Abstract (arXiv)

Despite the success in learning semantically meaningful, unsupervised disentangled representations, variational autoencoders (VAEs) and their variants face a fundamental theoretical challenge: substantial evidence indicates that unsupervised disentanglement is unattainable without implicit inductive bias, yet such bias remains elusive. In this work, we focus on exploring the implicit inductive bias that drive disentanglement in VAEs with factorization priors. By analyzing the total correlation in beta-TCVAE, we uncover a crucial implicit inductive bias called disentangling granularity, which leads to the discovery of an interesting "V"-shaped optimal Evidence Lower Bound (ELBO) trajectory within the parameter space. This finding is validated through over 100K experiments using factorized VAEs and our newly proposed model, beta-STCVAE. Notably, experimental results reveal that conventional factorized VAEs, constrained by fixed disentangling granularity, inherently tend to disentangle low-complexity feature. Whereas, appropriately tuning disentangling granularity, as enabled by beta-STCVAE, broadens the range of disentangled representations, allowing for the disentanglement of high-complexity features. Our findings unveil that disentangling granularity as an implicit inductive bias in factorized VAEs influence both disentanglement performance and the inference of the ELBO, offering fresh insights into the interpretability and inherent biases of VAEs.