Research map / Label-efficient and robust learning / Semi-supervised learning
Distributed Estimation on Semi-Supervised Generalized Linear Model
Jiyuan Tu, Weidong Liu, Xiaojun Mao
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 more local unlabeled samples (n* 100 -> 500) (ablation) | one-step SSDANE · Linear regression, m=100, n=100 | l2-error: 0.593 -> 0.161 | Improved |
| add more local unlabeled samples (n* 300 -> 900) (ablation) | one-step SSDANE · Logistic regression, m=100, n=300 | l2-error: 0.634 -> 0.285 | Improved |
| average local estimators from all machines (SSDANE-Avg(1)) instead of master only (ablation) | SSDANE · Linear regression, m=100, n=100, n*=100 | l2-error: 0.593 -> 0.178 | Improved |
| average local estimators from all machines (SSDANE-Avg(1)) instead of master only (ablation) | SSDANE · Logistic regression, m=100, n=300, n*=300 | l2-error: 0.634 -> 0.204 | Improved |
| increase averaging fraction from 0.2 to 1 (ablation) | SSDANE-Avg · Linear regression, m=100, n=100, n*=100 | l2-error: 0.199 -> 0.178 | Improved |
| increase number of machines m from 20 to 100 (ablation) | SSDANE-Avg(1) · Linear regression, n=100, n*=500 | l2-error: 0.142 -> 0.069 | Improved |
| run 5 steps instead of 1 step under random initialization (ablation) | SSDANE-Avg · Linear regression, m=100, n=100, n*=500 | l2-error: 0.307 -> 0.056 | Improved |
| run 5 steps instead of 1 step under random initialization with few unlabeled samples (ablation) | SSDANE · Linear regression, m=100, n=100, n*=100 | l2-error: 3.492 -> 33.780 | Hurt |
| use semi-supervised surrogate loss with 5 steps instead of naive semi-supervised loss Lss (ablation) | SSDANE · Linear regression, gradient covariance trace, m=100, n*=500 | rescaled trace of gradient covariance: 18.367 -> 0.998 | Improved |
| replace supervised DANE with semi-supervised SSDANE | distributed Newton · Linear and logistic regression | l2-error over iterations | Improved |
| add imputation of unlabeled data (DANEimp-Avg) instead of SSDANE-Avg | DANE-Avg · Linear and logistic regression, CelebA | l2-error / classification error | Hurt |
Closest papers by meaning
- Debiased Distributed Learning for Sparse Partial Linear Models in High Dimensions (JMLR 2022 · similarity 0.87)
- On Convergence of Distributed Approximate Newton Methods: Globalization, Sharper Bounds and Beyond (JMLR 2020 · similarity 0.85)
- Collaboratively Learning Linear Models with Structured Missing Data (NeurIPS 2023 · similarity 0.85)
- Semi-Supervised Sparse Gaussian Classification: Provable Benefits of Unlabeled Data (NeurIPS 2024 · similarity 0.84)
- Out-Of-Domain Unlabeled Data Improves Generalization (ICLR 2024 · similarity 0.83)
- Towards Realistic Model Selection for Semi-supervised Learning (ICML 2024 · 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.
