Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Learning under Latent Label Shift
Manley Roberts, Pranav Mani, Saurabh Garg, Zachary Chase Lipton
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 domain information via DDFA (SI) instead of SCAN alone | DDFA (SI) · CIFAR-20 r=20, alpha 0.5, kappa 8 | test accuracy: 0.445 -> 0.807 | Improved |
| Use DDFA (SI) instead of SCAN in the densest setting | DDFA (SI) · CIFAR-20 r=20, alpha 10, kappa 20 | test accuracy: 0.438 -> 0.355 | Hurt |
| Initialize DDFA backbone with SCAN weights (SI) instead of random (RI) (ablation) | DDFA · CIFAR-20 r=20, alpha 0.5, kappa 8 | test accuracy: 0.549 -> 0.807 | Improved |
| Initialize DDFA backbone with SCAN weights (SI) instead of random (RI) (ablation) | DDFA · CIFAR-20 r=20, alpha 10, kappa 20 | test accuracy: 0.203 -> 0.355 | Improved |
| Make label marginals denser (alpha 0.5 to alpha 10) | DDFA (SI) · CIFAR-20 r=20 | test accuracy: 0.807 -> 0.355 | Hurt |
| Increase number of domains r from 20 to 30 | DDFA (SI) · CIFAR-20, alpha 0.5, kappa 8 | test accuracy: 0.807 -> 0.868 | Improved |
| Increase number of domains r from 20 to 30 | DDFA (SI) · CIFAR-20, alpha 10, kappa 20 | test accuracy: 0.355 -> 0.517 | Improved |
| Add domain information via DDFA (SPI) instead of SCAN | DDFA (SPI) · FieldGuide-2 r=2, alpha 0.5, kappa 3 | test accuracy: 0.589 -> 0.947 | Improved |
| Use DDFA (SPI) instead of SCAN in the densest setting | DDFA (SPI) · FieldGuide-2 r=2, alpha 10, kappa 7 | test accuracy: 0.601 -> 0.676 | Improved |
| Use DDFA (SPI) instead of SCAN on label marginal error | DDFA (SPI) · FieldGuide-2 r=2, alpha 0.5, kappa 3 | Q_Y|D err: 0.880 -> 0.084 | Improved |
| Use DDFA (SPI) instead of SCAN on label marginal error with r=5 | DDFA (SPI) · FieldGuide-2 r=5, alpha 10, kappa 7 | Q_Y|D err: 0.226 -> 0.253 | Hurt |
| Use DDFA without SCAN on easy dataset | DDFA · CIFAR-10 | test accuracy | Hurt |
Closest papers by meaning
- Unsupervised Learning for Class Distribution Mismatch (ICML 2025 · similarity 0.85)
- Generalized Category Discovery under Domain Shift: A Frequency Domain Perspective (NeurIPS 2025 · similarity 0.85)
- Learning from Label Proportions: Bootstrapping Supervised Learners via Belief Propagation (ICLR 2024 · similarity 0.84)
- Disentangled Partial Label Learning (AAAI 2024 · similarity 0.84)
- Heterogeneous Label Shift: Theory and Algorithm (ICML 2025 · similarity 0.84)
- Semantic Feature Learning for Universal Unsupervised Cross-Domain Retrieval (NeurIPS 2024 · similarity 0.84)
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.
