Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Autoencoders Pursue PCA Directions (by Accident)
Check 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 |
|---|---|---|---|
| replace the diagonal posterior with a full-covariance posterior (ablation) | β-VAE · dSprites | DtO: 0.76 ± 0.08 -> 1.73 ± 0.14 | Hurt |
| replace the diagonal posterior with a full-covariance posterior (ablation) | β-VAE · Synth. Lin. | DtO: 0.00 ± 0.00 -> 0.34 ± 0.35 | Hurt |
| replace the diagonal posterior with a full-covariance posterior (ablation) | β-VAE · Synth. Non-Lin. | DtO: 0.18 ± 0.02 -> 0.55 ± 0.02 | Hurt |
| replace the diagonal posterior with a full-covariance posterior (ablation) | β-VAE · Synth. Non-Lin. | Disentanglement score: 0.73 ± 0.16 -> 0.42 ± 0.24 | Hurt |
| use a diagonal posterior rather than a full-covariance posterior (ablation) | β-VAE · dSprites | Disentanglement score: 0.12 ± 0.06 -> 0.33 ± 0.15 | Helped |
| use a diagonal posterior rather than a full-covariance posterior (ablation) | β-VAE · Synth. Lin. | Disentanglement score: 0.71 ± 0.31 -> 0.99 ± 0.01 | Mixed |
Closest papers by meaning
- Quantitative Understanding of VAE as a Non-linearly Scaled Isometric Embedding (ICML 2021 · similarity 0.92)
- Local Disentanglement in Variational Auto-Encoders Using Jacobian $L_1$ Regularization (NeurIPS 2021 · similarity 0.92)
- Embrace the Gap: VAEs Perform Independent Mechanism Analysis (NeurIPS 2022 · similarity 0.91)
- Why do Variational Autoencoders Really Promote Disentanglement? (ICML 2024 · similarity 0.91)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.90)
- Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders (ICML 2025 · 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.