Research map / Label-efficient and robust learning / Recognition and long-tail
Disentangling by Factorising
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 |
|---|---|---|---|
| Add a total-correlation penalty to the VAE objective | VAE · 2D Shapes | disentanglement score: around 0.73 -> around 0.82 | Helped |
| Add a total-correlation penalty to the VAE objective | VAE · 3D Shapes | reconstruction error: 3570 -> 3515 | Helped |
| Replace the product of code marginals with the factorial prior in the total-correlation objective (ablation) | FactorVAE · 2D Shapes | disentanglement performance | Hurt |
| Replace the discriminator-based density-ratio estimate with a batch estimate of the aggregate posterior (ablation) | FactorVAE · 2D Shapes | total-correlation estimate | Hurt |
| Use samples from the batch-estimated aggregate posterior instead of samples from the aggregate posterior (ablation) | FactorVAE · 2D Shapes | disentanglement-reconstruction trade-off | No clear effect |
| Use 5 continuous codes and no categorical code (ablation) | InfoWGAN-GP · 2D Shapes | disentanglement score | No clear effect |
| Use early stopping before training instability (ablation) | InfoGAN · 2D Shapes | disentanglement score | No clear effect |
| Increase the InfoGAN architecture size (ablation) | InfoGAN · 2D Shapes | disentanglement performance | No clear effect |
| Use a categorical latent code alongside continuous codes (ablation) | InfoWGAN-GP · 2D Shapes | disentanglement score | Hurt |
Closest papers by meaning
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.92)
- Variational Inference of Disentangled Latent Concepts from Unlabeled Observations (ICLR 2018 · similarity 0.92)
- Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations (ICML 2019 · similarity 0.91)
- Learning Disentangled Representation with Pairwise Independence (AAAI 2019 · similarity 0.91)
- Isolating Sources of Disentanglement in Variational Autoencoders (NeurIPS 2018 · similarity 0.90)
- 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.