Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hyperprior Induced Unsupervised Disentanglement of Latent Representations
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 CHyVAE instead of β-VAE or FactorVAE | VAE-based unsupervised disentanglement · 2DShapes continuous factors | reconstruction error | Helped |
| Use CHyVAE instead of β-VAE or FactorVAE | VAE-based unsupervised disentanglement · 2DShapes continuous factors | disentanglement | Helped |
| Use CHyVAE instead of β-VAE or FactorVAE | VAE-based unsupervised disentanglement · 2DShapes including shape | disentanglement metric: 0.909; 0.905 | Mixed |
| Use CHyVAE instead of β-VAE or FactorVAE | VAE-based unsupervised disentanglement · CorrelatedEllipses | disentanglement metric | Helped |
| Use CHyVAE instead of β-VAE or FactorVAE | VAE-based unsupervised disentanglement · CorrelatedEllipses | reconstruction error | Helped |
| Use CHyVAE instead of β-VAE | VAE-based unsupervised disentanglement · 3DFaces, 3DChairs, and CelebA | reconstruction error | Helped |
| Use CHyVAE instead of FactorVAE | VAE-based unsupervised disentanglement · 3DFaces and CelebA | qualitative disentanglement | Mixed |
| Use CHyVAE instead of FactorVAE | VAE-based unsupervised disentanglement · 3DChairs | leg-style factor discovery | No clear effect |
| Increase the inverse-Wishart degrees of freedom ν (ablation) | CHyVAE · 2DShapes | disentanglement | Helped |
| Decrease the inverse-Wishart degrees of freedom ν, allowing more deviation from identity covariance (ablation) | CHyVAE · CorrelatedEllipses | disentanglement | Helped |
| Decrease the inverse-Wishart degrees of freedom ν, allowing more deviation from identity covariance (ablation) | CHyVAE · CorrelatedEllipses | reconstruction error | Helped |
| Replace the structured variational posterior with a mean-field factorization (ablation) | CHyVAE · preliminary experiments | training stability | Hurt |
Closest papers by meaning
- Disentanglement via Latent Quantization (NeurIPS 2023 · similarity 0.91)
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.91)
- When Is Unsupervised Disentanglement Possible? (NeurIPS 2021 · similarity 0.91)
- Variational Inference of Disentangled Latent Concepts from Unlabeled Observations (ICLR 2018 · similarity 0.90)
- Orthogonality-Enforced Latent Space in Autoencoders: An Approach to Learning Disentangled Representations (ICML 2023 · similarity 0.90)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · similarity 0.89)
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.