Research map / Probabilistic and scientific modeling / Molecules and proteins
Learning disentangled representations via product manifold projection
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 |
|---|---|---|---|
| Use product-manifold projection instead of Ada-GVAE | Disentangled representation learning · Shapes3D | MIG: 56.2% -> 63.7% | Helped |
| Use product-manifold projection instead of Ada-GVAE | Disentangled representation learning · DSprites | MIG: 26.6% -> 25.7% | Hurt |
| Use product-manifold projection instead of Ada-GVAE | Disentangled representation learning · Cars3D | MIG: 15.0% -> 25.9% | Helped |
| Use product-manifold projection instead of Ada-GVAE | Disentangled representation learning · SmallNORB | FactorVAE: 55.5% -> 88.2% | Helped |
| Allow a variable number of factors to change between paired samples | Ours · Shapes3D | MIG: 63.7% -> 62.3% | Hurt |
| Allow a variable number of factors to change between paired samples | Ours · Shapes3D | BetaVAE: 100% -> 98.9% | Hurt |
| Remove the consistency loss (ablation) | Ours · Shapes3D | disentanglement scores | Hurt |
| Remove the regularization loss (ablation) | Ours · Shapes3D | disentanglement scores | Hurt |
| Remove both the consistency and regularization losses (ablation) | Ours · Shapes3D | disentanglement scores | Hurt |
| Increase latent-space dimensionality on FAUST | Ours · FAUST | MIG-KM: 10.6% -> 10.3% -> 9.5% | No clear effect |
| Increase latent-space dimensionality on FAUST | Ours · FAUST | Reconstruction error: 2.56e-3 -> 2.04e-3 -> 1.24e-3 | Helped |
| Use multidimensional latent subspaces rather than a one-dimensional latent space | Ours · FAUST | Reconstruction error: 2.56e-3 -> 1.26e-3 | Helped |
Closest papers by meaning
- Evaluating the Disentanglement of Deep Generative Models through Manifold Topology (ICLR 2021 · similarity 0.91)
- Disentangling by Factorising (ICML 2018 · similarity 0.89)
- When Is Unsupervised Disentanglement Possible? (NeurIPS 2021 · similarity 0.89)
- Variational Inference of Disentangled Latent Concepts from Unlabeled Observations (ICLR 2018 · similarity 0.88)
- Disentangled Representation Learning with the Gromov-Monge Gap (ICLR 2025 · similarity 0.87)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.87)
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.