Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
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 |
|---|---|---|---|
| Add covariance regularization on inferred posterior means | DIP-VAE-I · 2D Shapes test | Z-diff: 81.3 -> 98.7 | Helped |
| Add covariance regularization on inferred posterior means | DIP-VAE-I · 2D Shapes test | Reconstruction error (per pixel): 0.0017 -> 0.0018 | No clear effect |
| Use stronger β regularization (ablation) | β-VAE · 2D Shapes test | Reconstruction error (per pixel): 0.0032 -> 0.0113 | Hurt |
| Regularize covariance of the inferred prior, including posterior variance, rather than only posterior means (ablation) | DIP-VAE-I, 2D Shapes · 2D Shapes test | SAP: 0.1889 -> 0.5253 | Helped |
| Increase covariance regularization strength (ablation) | DIP-VAE-II · 2D Shapes test | SAP: 0.2188 -> 0.5253 | Helped |
| Increase covariance regularization strength (ablation) | DIP-VAE-II · 2D Shapes test | Reconstruction error (per pixel): 0.0023 -> 0.0079 | Hurt |
| Add covariance regularization on inferred posterior means | DIP-VAE-I · CelebA test | SAP: 0.35 -> 3.69 | Helped |
| Add covariance regularization on inferred posterior means | DIP-VAE-I · CelebA test | Reconstruction error (per pixel): 0.0876 -> 0.0904 | No clear effect |
| Increase covariance regularization strength (ablation) | DIP-VAE-II · CelebA test | SAP: 2.94 -> 3.93 | Helped |
| Increase covariance regularization strength (ablation) | DIP-VAE-II · CelebA test | Reconstruction error (per pixel): 0.0884 -> 0.1477 | Hurt |
| Use β-VAE instead of DIP-VAE-I | DIP-VAE-I · CelebA test | Reconstruction error (per pixel): 0.0904 -> 0.1572 | Hurt |
Closest papers by meaning
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.92)
- Disentangling by Factorising (ICML 2018 · similarity 0.92)
- Learning Disentangled Joint Continuous and Discrete Representations (NeurIPS 2018 · similarity 0.91)
- Unsupervised Disentanglement Without Compromises : How Functional Orthogonality Enforces Identifiability (ICML 2026 · similarity 0.91)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.91)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · 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.