Research map / Learning theory and algorithms / Matrix and tensor methods
Disentangling by Subspace Diffusion
David Pfau, Irina Higgins, Alex Botev, Sébastien Racanière
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Apply a random orthogonal rotation to the true latent state (ablation) | GEOMANCER · Stanford Bunny | average angle: 0.024 -> 0.024 | No clear effect |
| Randomly scale individual latent dimensions (ablation) | GEOMANCER · Stanford Bunny | average angle: 0.024 -> 0.72±0.68 | Hurt |
| Apply a random linear transformation to the true latent state (ablation) | GEOMANCER · Stanford Bunny | average angle: 0.024 -> 1.42±0.09 | Hurt |
| Use a Laplacian Eigenmaps embedding instead of the true latent state (ablation) | GEOMANCER · Stanford Bunny | average angle: 0.024 -> 0.37; Chance 1.26±0.23 | Mixed |
| Use pixels as input instead of the true latent state (ablation) | GEOMANCER · Stanford Bunny | average angle: 1.25; Chance 1.26±0.23 | No clear effect |
| Use β-VAE latent vectors instead of the true latent state (ablation) | GEOMANCER · Stanford Bunny | average angle: 1.30±0.02; Chance 1.26±0.23 | No clear effect |
| Apply a random orthogonal rotation to the true latent state (ablation) | GEOMANCER · Stanford Dragon | average angle: 0.023 -> 0.023 | No clear effect |
| Randomly scale individual latent dimensions (ablation) | GEOMANCER · Stanford Dragon | average angle: 0.023 -> 0.96±0.70 | Hurt |
| Apply a random linear transformation to the true latent state (ablation) | GEOMANCER · Stanford Dragon | average angle: 0.023 -> 1.26±0.37 | Hurt |
| Use a Laplacian Eigenmaps embedding instead of the true latent state (ablation) | GEOMANCER · Stanford Dragon | average angle: 0.023 -> 0.32; Chance 1.26±0.23 | Mixed |
| Use pixels as input instead of the true latent state (ablation) | GEOMANCER · Stanford Dragon | average angle: 1.26; Chance 1.26±0.23 | No clear effect |
| Use β-VAE latent vectors instead of the true latent state (ablation) | GEOMANCER · Stanford Dragon | average angle: 1.15±0.09; Chance 1.26±0.23 | No clear effect |
Closest papers by meaning
- Learning disentangled representations via product manifold projection (ICML 2021 · similarity 0.88)
- Efficient Diffusion Models for Symmetric Manifolds (ICML 2025 · similarity 0.86)
- Manifold Diffusion Fields (ICLR 2024 · similarity 0.85)
- Generative Modeling on Manifolds Through Mixture of Riemannian Diffusion Processes (ICML 2024 · similarity 0.85)
- Geometric Analysis of Nonlinear Manifold Clustering (NeurIPS 2024 · similarity 0.84)
- Finsler Multi-Dimensional Scaling: Manifold Learning for Asymmetric Dimensionality Reduction and Embedding (CVPR 2025 · similarity 0.84)
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.
