Research map / Label-efficient and robust learning / Recognition and long-tail
Theory and Evaluation Metrics for Learning Disentangled Representations
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 FactorVAE instead of β-VAE or AAE | Representation learning · dSprites scale, Pos X, and Pos Y | RMIG: Scale: 0.1533 -> 0.7139 (β-VAE); Pos X: 0.4127 -> 0.7198 (β-VAE); Pos Y: 0.4193 -> 0.7256 (β-VAE) | Helped |
| Use FactorVAE instead of β-VAE or AAE | Representation learning · dSprites | JEMMIG: Shape: 0.8642 -> 0.6841 (β-VAE); Scale: 0.8087 -> 0.3422 (β-VAE); Rotation: 0.9199 -> 0.7204 (β-VAE); Pos X: 0.5629 -> 0 | Helped |
| Increase β in β-VAE | β-VAE · FactorVAE and β-VAE experiments | informativeness | Hurt |
| Increase the FactorVAE TC-loss weight | FactorVAE · FactorVAE experiments | I(z_i,z_≠i) | No clear effect |
| Set β=10 | β-VAE · FactorVAE and β-VAE experiments | WSEPIN and SEPIN@3 | Helped |
| Use FactorVAE instead of β-VAE | Representation learning · CelebA | interpretability (RMIG and JEMMIG) | Helped |
| Use more than 200 quantization bins (ablation) | RMIG evaluation · Metric bin-count sensitivity | RMIG stability | No clear effect |
| Increase the quantization bin count (ablation) | JEMMIG evaluation · Metric bin-count sensitivity | JEMMIG | Hurt |
| Increase the number of samples used for estimation (ablation) | Sampling-based metrics · dSprites | JEMMIG and WSEPIN | No clear effect |
| Increase the number of latent variables (ablation) | β-VAE with β=10 · High-dimensional latent evaluation | WSEPIN stability | Hurt |
Closest papers by meaning
- Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations (ICML 2019 · similarity 0.93)
- A Framework for the Quantitative Evaluation of Disentangled Representations (ICLR 2018 · similarity 0.92)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (NeurIPS 2024 · similarity 0.91)
- Revisiting Disentanglement in Downstream Tasks: A Study on Its Necessity for Abstract Visual Reasoning (AAAI 2024 · similarity 0.90)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.89)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · 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.