Research map / Probabilistic and scientific modeling / Variational and generative sampling
Exploring the Latent Space of Autoencoders with Interventional Assays
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 |
|---|---|---|---|
| Increase β from 1 to 8 (ablation) | β-VAE · 3D-Shapes (d=12) | CDS: 0.44 -> 0.71 | Helped |
| Increase β from 1 to 8 (ablation) | β-VAE · 3D-Shapes (d=24) | CDS: 0.52 -> 0.78 | Helped |
| Increase β from 1 to 8 (ablation) | β-VAE · 3D-Shapes | DCI-D: 0.3 -> 0.66 (d=12); 0.49 -> 0.73 (d=24) | Helped |
| Increase β from 1 to 8 (ablation) | β-VAE · 3D-Shapes | IRS: 0.44 -> 0.63 (d=12); 0.48 -> 0.64 (d=24) | Helped |
| Increase β from 1 to 8 (ablation) | β-VAE · 3D-Shapes | MIG: 0.07 -> 0.21 (d=12); 0.09 -> 0.2 (d=24) | Helped |
| Increase β from 1 to 8 (ablation) | β-VAE · MPI3D Toy | CDS: 0.69 -> 1 | Mixed |
| Increase β from 1 to 8 (ablation) | β-VAE · MPI3D Toy | DCI-D: 0.33 -> 0.13 | Hurt |
| Increase β from 1 to 8 (ablation) | β-VAE · MPI3D Real | CDS: 0.61 -> 0.7 | Mixed |
| Increase β from 1 to 8 (ablation) | β-VAE · MPI3D Real | DCI-D: 0.24 -> 0.08 | Hurt |
| Use conditioned causal responses instead of encoder-based correlations to assess disentanglement | CDS versus DCI-D · 3D-Shapes | disentanglement score: 4-VAE, d=24: CDS 0.72; DCI-D 0.67 | Mixed |
| Follow a high-curvature response-map path rather than the shortest Euclidean path | VAE latent interpolation · Double helix toy dataset | interpolation quality | Helped |
Closest papers by meaning
- Disentanglement via Latent Quantization (NeurIPS 2023 · similarity 0.90)
- Variational Autoencoders with Jointly Optimized Latent Dependency Structure (ICLR 2019 · similarity 0.90)
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.90)
- On Incorporating Inductive Biases into VAEs (ICLR 2022 · similarity 0.90)
- Shape your Space: A Gaussian Mixture Regularization Approach to Deterministic Autoencoders (NeurIPS 2021 · similarity 0.89)
- Masked Autoencoders Are Effective Tokenizers for Diffusion Models (ICML 2025 · similarity 0.88)
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.