Research map / Label-efficient and robust learning / Domain adaptation and generalization
ELSA: Efficient Label Shift Adaptation through the Lens of Semiparametric Models
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 ELSA moment-matching estimator instead of MLLS without calibration | ELSA vs MLLS · MNIST Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 3.156E-03 -> 9.317E-04 | Helped |
| Use ELSA instead of uncalibrated MLLS on CIFAR-10 | ELSA vs MLLS · CIFAR-10 Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 2.565E-03 -> 7.279E-04 | Helped |
| Use ELSA instead of best calibrated MLLS on CIFAR-100 | ELSA vs MLLS-BCTS · CIFAR-100 Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 0.133 (MLLS-BCTS) vs 0.207 (ELSA) | Hurt |
| Remove post-prediction calibration from ELSA (none vs BCTS/NBVS/TS/VS) (ablation) | ELSA · MNIST Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 9.317E-04 (none) vs 8.419E-04 (BCTS) and 8.676E-04 (NBVS) | No clear effect |
| Calibration (BCTS, VS) applied to MLLS (ablation) | MLLS · MNIST Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 3.156E-03 (none) -> 1.016E-03 (BCTS) | Helped |
| Calibration applied to MLLS on CIFAR-10 (ablation) | MLLS · CIFAR-10 Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 2.565E-03 (none) -> 7.503E-04 (BCTS) | Helped |
| Use ELSA instead of BBSE-soft/RLLS-soft on MNIST Dirichlet | ELSA vs BBSE/RLLS · MNIST Dirichlet (n=4500, alpha=1.0) | MSE trimmed 5%: 1.335E-03 (RLLS-soft) -> 9.317E-04 | Helped |
| Use ELSA instead of uncalibrated MLLS under tweakone shift | ELSA vs MLLS · MNIST tweakone (n=4500, rho=0.9) | MSE trimmed 5%: ELSA 8.473E-04 vs MLLS-VS 4.348E-04 (calibrated MLLS better) | Mixed |
Closest papers by meaning
- RLSbench: Domain Adaptation Under Relaxed Label Shift (ICML 2023 · similarity 0.91)
- Pseudo-Labeling for Unsupervised Domain Adaptation with Kernel GLMs (NeurIPS 2026 · similarity 0.88)
- Label Shift Meets Online Learning: Ensuring Consistent Adaptation with Universal Dynamic Regret (CVPR 2025 · similarity 0.88)
- Enhancing Semi-supervised Domain Adaptation via Effective Target Labeling (AAAI 2024 · similarity 0.86)
- Online Feature Updates Improve Online (Generalized) Label Shift Adaptation (NeurIPS 2024 · similarity 0.86)
- Heterogeneous Label Shift: Theory and Algorithm (ICML 2025 · 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.