Research map / Label-efficient and robust learning / Recognition and long-tail
On the Transfer of Disentangled Representations in Realistic Settings
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 weak supervision with pairs differing in one factor | β-VAE representation learning · Robotic finger dataset | disentanglement: 29% of them have >99% DCI, some of them up to 99.89% | Helped |
| Add Gaussian noise to the input during training (ablation) | β-VAE · OOD2, including sim2real | transfer error | Helped |
| Add Gaussian noise to the input during training (ablation) | β-VAE · OOD1 | transfer error | No clear effect |
| Increase representation disentanglement | Downstream GBT predictors · OOD1-A, OOD1-B, and OOD1-C | transfer error | Helped |
| Increase representation disentanglement | Downstream MLP predictors · OOD1 | transfer error | Mixed |
| Increase representation disentanglement | Downstream GBT predictors · OOD2-A simulated held-out colors | transfer error | Helped |
| Increase representation disentanglement | Downstream predictors · OOD2-B real-world images | transfer error | No clear effect |
| Use DCI or MIG to assess disentanglement | Representation model selection · Robotic finger dataset | correlation with downstream GBT test accuracy | Helped |
| Use SAP or Modularity as the disentanglement metric | Representation model selection · Robotic finger dataset | disentanglement measurement | No clear effect |
| Add batch normalization, layer normalization, or dropout (ablation) | VAE architecture · Robotic finger dataset | ELBO, sample quality, and disentanglement | No clear effect |
| Add layer normalization before posterior parameterization (ablation) | VAE encoder · Robotic finger dataset | training stability | Helped |
| Initialize residual blocks as identity using a learnable scalar gate | Residual VAE architecture · Robotic finger dataset | training stability | Helped |
Closest papers by meaning
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.93)
- Revisiting Disentanglement in Downstream Tasks: A Study on Its Necessity for Abstract Visual Reasoning (AAAI 2024 · similarity 0.90)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.89)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (NeurIPS 2024 · similarity 0.89)
- Disentangled Representation Learning with the Gromov-Monge Gap (ICLR 2025 · 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.