Research map / Probabilistic and scientific modeling / Molecules and proteins
Commutative Lie Group VAE for Disentanglement Learning
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 an adaptive commutative Lie group representation instead of a vector-space latent | Commutative Lie Group VAE · DSprites | FactorVAE metric: 82.15 ± 0.88 -> 86.1 ± 2.0 | Helped |
| Use an adaptive commutative Lie group representation instead of a vector-space latent | Commutative Lie Group VAE · 3DShapes | FactorVAE metric: 91 -> 93.2 ± 4.0 | Helped |
| Add a shared encoder-decoder feature constraint (ablation) | VAE on DSprites test · DSprites test | FactorVAE metric: 69.4 ± 10.9 -> 74.6 ± 8.1 | Helped |
| Add an exponential mapping group layer (ablation) | Bottleneck-VAE on DSprites test · DSprites test | FactorVAE metric: 74.6 ± 8.1 -> 83.6 ± 3.2 | Helped |
| Increase group representation size from 4 to 100 (ablation) | Lie Group VAE on DSprites test · DSprites test | FactorVAE metric: 23.6 ± 3.3 -> 85.5 ± 2.2 | Helped |
| Vary one-parameter subgroup decomposition weight; weight 40 gives the reported SAP peak (ablation) | Commutative Lie Group VAE on DSprites test · DSprites test | SAP: 40.7 ± 12.2 -> 50.8 ± 5.0; weight 80: 47.1 ± 8.6 | Mixed |
| Vary Hessian penalty weight; weight 20 gives the reported SAP peak (ablation) | Commutative Lie Group VAE on DSprites test · DSprites test | SAP: 40.7 ± 12.2 -> 54.1 ± 1.2; weight 40: 48.2 ± 1.9 | Mixed |
| Use Hessian penalty rather than one-parameter decomposition at their best reported settings (ablation) | Commutative Lie Group VAE on DSprites test · DSprites test | SAP: 50.8 ± 5.0 -> 54.1 ± 1.2 | Helped |
Closest papers by meaning
- Towards Building A Group-based Unsupervised Representation Disentanglement Framework (ICLR 2022 · similarity 0.91)
- Disentangling by Factorising (ICML 2018 · similarity 0.88)
- Disentanglement Analysis with Partial Information Decomposition (ICLR 2022 · similarity 0.88)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · similarity 0.88)
- Lie Group Decompositions for Equivariant Neural Networks (ICLR 2024 · similarity 0.87)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (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.