Research map / Probabilistic and scientific modeling / Variational and generative sampling
Disentanglement Analysis with Partial Information Decomposition
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 UniBound instead of MIG to evaluate representations with injected redundant information | disentanglement metrics · Gaussian toy model, redundancy attack | response to injected redundancy: MIG \to \frac12 \log \Paren{1 + \frac{1 - (1 - \frac2K)^2}{\sigma^2}}; \textsc{UniBound} \to 0 | Helped |
| Inject redundant information by appending noisy mixed copies of the representation (ablation) | β-VAE and β-TCVAE representations · dSprites and 3dshapes | UniBound | Hurt |
| Inject redundant information by appending noisy mixed copies of the representation (ablation) | BetaVAE metric and FactorVAE metric · dSprites and 3dshapes | disentanglement score | No clear effect |
| Inject redundant information by appending noisy mixed copies of the representation (ablation) | MIG · dSprites and 3dshapes | MIG | Mixed |
| Use FactorVAE rather than β-VAE | VAE representations · dSprites | MIG | Helped |
| Use FactorVAE rather than β-VAE | VAE representations · dSprites | UniBound | Hurt |
| Use β-TCVAE | VAE representations · dSprites scale, position x, and position y | unique information | Helped |
| Use β-TCVAE | VAE representations · dSprites orientation | synergistic information | Hurt |
| Introduce a discrete latent variable | JointVAE · dSprites shape and orientation; 3dshapes shape and orientation | unique information | Helped |
| Increase regularization strength (β for β-VAE and β-TCVAE, γ for FactorVAE, or continuous capacity for JointVAE) (ablation) | VAE variants · dSprites and 3dshapes | UniBound | Helped |
| Reduce JointVAE continuous-variable capacity (ablation) | JointVAE · dSprites and 3dshapes | redundancy | Helped |
Closest papers by meaning
- Theory and Evaluation Metrics for Learning Disentangled Representations (ICLR 2020 · similarity 0.91)
- Variational Inference of Disentangled Latent Concepts from Unlabeled Observations (ICLR 2018 · similarity 0.91)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · similarity 0.90)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (NeurIPS 2024 · similarity 0.90)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.89)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.89)
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.