Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-supervised Inference for Block-wise Missing Data without Imputation
Shanshan Song, Yuanyuan Lin, Yong Zhou
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 |
|---|---|---|---|
| use the proposed imputation-free estimator instead of imputation-based X. et al. | block-missing semi-supervised inference · Simulation E2(i) (n=300, N(0,1)) | Size of test for G2: 0.160 -> 0.070 | Improved |
| avoid imputation of missing blocks | block-missing semi-supervised inference · Simulation E2(ii) (n=300) | computation time (minutes): 37.522 -> 1.265 | Improved |
| use incomplete labelled data and unlabelled data rather than complete cases only (Z. & Z.) | debiased lasso inference · Simulation E3 (n4 = 60) | AL over S0: 1.209 -> 0.528 | Improved |
| use proposed method rather than complete-case van de Geer debiasing | debiased lasso inference · Simulation E3 (n4 = 200) | AL over S0: 0.298 -> 0.411 | Hurt |
| use proposed method rather than van de Geer debiasing with complete cases | debiased lasso inference · Simulation E3 (n4 = 60) | Size of test for G2: 0.100 -> 0.050 | Improved |
| use proposed method rather than complete-case Z. & Z. | debiased lasso inference · Simulation E3 (n4 = 60) | ACP over S0: 0.938 -> 0.884 | Hurt |
| increase labelled sample size n from 200 to 300 | proposed method · Simulation E1 (N=5000, N(0,1)) | AL over S0: 0.897 -> 0.728 | Improved |
| increase labelled sample size n from 300 to 600 | proposed method · Simulation E2(i) (N(0,1)) | AL over S0: 0.623 -> 0.449 | Improved |
| increase unlabelled sample size N from 1000 to 5000 | proposed method · Simulation E1 (n=300, N(0,1)) | ACP over S0: 0.872 -> 0.907 | Improved |
| heavier-tailed t(3) errors instead of N(0,1) | proposed method · Simulation E1 (n=300, N=5000) | AL over S0: 0.728 -> 0.906 | Hurt |
| use proposed test instead of complete-case van de Geer test | hippocampus biomarker set G1 · ADNI MMSE | p-value: 3.048×10−2 -> 6.680×10−5 | Improved |
| use proposed test instead of van de Geer test | entorhinal biomarker set G2 · ADNI MMSE | p-value: 4.028×10−5 -> 1.785×10−4 | No clear effect |
Closest papers by meaning
- Are labels informative in semi-supervised learning? Estimating and leveraging the missing-data mechanism. (ICML 2023 · similarity 0.87)
- Language Model Augmented Semi-Supervised Statistical Inference (ICML 2026 · similarity 0.86)
- Naive imputation implicitly regularizes high-dimensional linear models (ICML 2023 · similarity 0.85)
- Random features models: a way to study the success of naive imputation (ICML 2024 · similarity 0.84)
- Resilient Semi-Supervised Inference with Heterogeneous Unlabeled Data (NeurIPS 2026 · similarity 0.84)
- DiffPuter: Empowering Diffusion Models for Missing Data Imputation (ICLR 2025 · similarity 0.83)
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.
