Research map / Label-efficient and robust learning / Recognition and long-tail
Disentangling Factors of Variations Using Few Labels
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 supervised regularizer during training instead of using labels only for model selection | VAE disentanglement methods · Four disentanglement datasets; 100 labeled examples | win rate over U/S on a random test metric and dataset: 70.5% | Helped |
| Increase labeled examples from 100 to 1000 for semi-supervised training | S²/S methods · Four disentanglement datasets | win rate over U/S: 70.5% -> 80.6% | Helped |
| Use coarse binned labels for supervised validation | U/S model selection · Four disentanglement datasets | rank correlation with test disentanglement metrics | No clear effect |
| Use labels with 10% random corruption | U/S model selection · Four disentanglement datasets | rank correlation with test disentanglement metrics | No clear effect |
| Train with labels for only two randomly selected factors | S²/S methods · Four disentanglement datasets; 100 labels | win rate over U/S: 50.0% | Mixed |
| Use binary cross-entropy rather than L2 as the supervised loss (ablation) | S²/S methods · Four disentanglement datasets | disentanglement performance | Helped |
| Randomly permute the ordering of factor labels (ablation) | S²/S methods · Four disentanglement datasets; MIG, DCI, and SAP | win rate over U/S: 66.1% -> 33.9% | Hurt |
| Replace perfect labels with binned labels during S²/S training (ablation) | S²/S methods · Four disentanglement datasets; random test metric and dataset | win rate over U/S: 79.5% -> 79.2% | No clear effect |
| Replace perfect labels with noisy labels during S²/S training (ablation) | S²/S methods · Four disentanglement datasets; random test metric and dataset | win rate over U/S: 79.5% -> 81.3% | No clear effect |
Closest papers by meaning
- Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations (ICML 2019 · similarity 0.93)
- On Disentangled Representations Learned from Correlated Data (ICML 2021 · similarity 0.91)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.88)
- Disentangled Representation Learning in Non-Markovian Causal Systems (NeurIPS 2024 · similarity 0.88)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · similarity 0.88)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.88)
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.