Research map / Probabilistic and scientific modeling / Variational and generative sampling
Isolating Sources of Disentanglement in Variational Autoencoders
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use β-TCVAE instead of β-VAE | VAE disentanglement training · dSprites and 3D Faces | MIG | Helped |
| Use β-TCVAE instead of InfoGAN | VAE disentanglement training · dSprites and 3D Faces | MIG | Helped |
| Use β-TCVAE instead of FactorVAE | VAE disentanglement training · dSprites and 3D Faces | MIG | Mixed |
| Remove the index-code MI penalty (α=0) (ablation) | β-TCVAE · dSprites and 3D Faces | MIG | No clear effect |
| Use a factorial normalizing flow prior (ablation) | β-TCVAE · dSprites and 3D Faces | MIG | No clear effect |
| Use lower batch sizes (ablation) | β-TCVAE · 3D Faces | MIG | No clear effect |
| Use minibatch stratified sampling (MSS) instead of minibatch-weighted sampling (MWS) (ablation) | β-TCVAE estimator · dSprites and 3D Faces experiments | performance | No clear effect |
| Increase the TC penalty while retaining α=γ=1 | β-TCVAE · dSprites and 3D Faces | MIG | Helped |
Closest papers by meaning
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.94)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · similarity 0.91)
- Orthogonality-Enforced Latent Space in Autoencoders: An Approach to Learning Disentangled Representations (ICML 2023 · similarity 0.91)
- Disentanglement Analysis with Partial Information Decomposition (ICLR 2022 · similarity 0.91)
- Demystifying Inductive Biases for (Beta-)VAE Based Architectures (ICML 2021 · similarity 0.91)
- Disentangling by Factorising (ICML 2018 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.