Research map / Probabilistic and scientific modeling / Variational and generative sampling
Topological Obstructions and How to Avoid Them
Babak Esmaeili, Robin Walters, Heiko Zimmermann, Jan-Willem van de Meent
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 |
|---|---|---|---|
| replace unimodal Gaussian VAE posterior with a normalizing-flow posterior (GF-VAE, beta=4) | VAE (beta=4) · L-shaped Tetromino SO(2) | # homeomorphic encoders: 2/15 -> 9/15 | Improved |
| replace VAE (beta=1) posterior with flow posterior | VAE (beta=1) · L-shaped Tetromino SO(2) | # homeomorphic encoders: 0/15 -> 5/15 | Improved |
| replace VAE (beta=4) posterior with flow posterior | VAE (beta=4) · Airplane SO(2) | continuity error: 67.88 \pm 59.20 -> 27.70 \pm 29.23 | Improved |
| replace VAE (beta=4) posterior with flow posterior | VAE (beta=4) · Teapot SO(2) | # homeomorphic encoders: 15/15 -> 13/15 | Hurt |
| add action-decoder to GF-VAE (beta=4) | GF-VAE (beta=4) · L-shaped Tetromino SO(2) | continuity error: 24.04 \pm 29.14 -> 19.35 \pm 23.37 | Improved |
| add action-decoder to GF-VAE (beta=4) | GF-VAE (beta=4) · Airplane SO(2) | continuity error: 27.70 \pm 29.23 -> 33.26 \pm 40.03 | Hurt |
| increase KL weight beta from 1 to 4 (ablation) | GF-VAE · Teapot SO(2) | # homeomorphic encoders: 7/15 -> 13/15 | Improved |
| increase KL weight beta from 1 to 4 (ablation) | VAE · Teapot SO(2) | # homeomorphic encoders: 6/15 -> 15/15 | Improved |
| increase KL weight beta from 1 to 4 (ablation) | VAE · L-shaped Tetromino SO(2) | # homeomorphic encoders: 0/15 -> 2/15 | Improved |
| add y-space regularization (||y||-1)^2 to VAE (ablation) | VAE · Teapot SO(2) | # homeomorphic encoders: 6/15 -> 1/15 | Hurt |
| add y-space regularization to VAE (ablation) | VAE · Airplane SO(2) | # homeomorphic encoders: 0/15 -> 3/15 | No clear effect |
| add supervision of ground-truth representations to VAE | VAE · Airplane SO(2) | # homeomorphic encoders: 0/15 | No clear effect |
Closest papers by meaning
- Graph Geometry-Preserving Autoencoders (ICML 2024 · similarity 0.87)
- Diffeomorphic interpolation for efficient persistence-based topological optimization (NeurIPS 2024 · similarity 0.87)
- Towards Scalable Topological Regularizers (ICLR 2025 · similarity 0.87)
- The Geometric Wall: Manifold Structure Predicts Layerwise Sparse Autoencoder Scaling Laws (NeurIPS 2026 · similarity 0.87)
- Connecting Neural Models Latent Geometries with Relative Geodesic Representations (NeurIPS 2025 · similarity 0.86)
- What Secrets Do Your Manifolds Hold? Understanding the Local Geometry of Generative Models (ICLR 2025 · similarity 0.86)
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.
