Research map / Label-efficient and robust learning / Recognition and long-tail
DisUnknown: Distilling Unknown Factors for Disentanglement Learning
Sitao Xiang, Yuming Gu, Pengda Xiang, Menglei Chai, Hao Li, Yajie Zhao, Mingming He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 unknown distillation (encoded unknown code plus feature matching loss) (ablation) | DisUnknown · 3D Shapes, unknown = floor hue | consistency ratio: 63.42% -> 100.00% | Improved |
| add unknown distillation (encoded unknown code plus feature matching loss) (ablation) | DisUnknown · 3D Shapes, unknown = wall hue | consistency ratio: 55.63% -> 100.00% | Improved |
| add unknown distillation (encoded unknown code plus feature matching loss) (ablation) | DisUnknown · 3D Shapes, unknown = object hue | consistency ratio: 68.76% -> 100.00% | Improved |
| use adversarial Stage II classifier instead of non-adversarial (ablation) | DisUnknown · MNIST | MIG: 0.916 -> 0.978 | Improved |
| use adversarial Stage II classifier instead of non-adversarial (ablation) | DisUnknown · MNIST | MSE: 0.0103 -> 0.0086 | Improved |
| add unknown code condition / code distance term (ablation) | DisUnknown · MNIST | MIG: 0.930 -> 0.978 | Improved |
| add unknown code condition / code distance term (ablation) | DisUnknown · MNIST | MSE: 0.0402 -> 0.0086 | Improved |
| add Negative Log Unlikelihood and sample-space classification in Stage I (ablation) | DisUnknown · 3D Chairs, rotation unknown | code-space fluctuation | Improved |
| merge more labeled factors into the unknown factor (ablation) | DisUnknown · 3D Shapes, object hue | MIG: 0.9501 -> 0.9868 | No clear effect |
| choose different factor as unknown (ablation) | DisUnknown · 3D Shapes | MIG: 0.9411 -> 0.9825 | No clear effect |
| add weak supervision (labeled factor) vs unsupervised methods | DisUnknown · MNIST | MIG: 0.568 -> 0.978 | Improved |
Closest papers by meaning
- Disentangling Factors of Variations Using Few Labels (ICLR 2020 · similarity 0.89)
- Learning Discrete and Continuous Factors of Data via Alternating Disentanglement (ICML 2019 · similarity 0.89)
- Weakly Supervised Disentanglement with Guarantees (ICLR 2020 · similarity 0.88)
- Weakly-Supervised Disentanglement Without Compromises (ICML 2020 · similarity 0.88)
- Disentanglement Beyond Static vs. Dynamic: A Benchmark and Evaluation Framework for Multi-Factor Sequential Representations (NeurIPS 2025 · similarity 0.88)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.86)
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.
