Research map / Label-efficient and robust learning / Semi-supervised learning
How unlabeled data improve generalization in self-training? A one-hidden-layer theoretical analysis
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 |
|---|---|---|---|
| Increase the number of unlabeled samples (ablation) | Iterative self-training · Synthetic Gaussian data | relative weight error: linear function of 1/sqrt(M) | Helped |
| Increase the number of unlabeled samples (ablation) | Iterative self-training · Synthetic Gaussian data | convergence rate: linear function of 1/sqrt(M) | Helped |
| Increase the number of labeled samples (ablation) | Iterative self-training · Synthetic Gaussian data | relative weight error | Helped |
| Increase the variance of unlabeled inputs above 1 (ablation) | Iterative self-training · Synthetic Gaussian data | convergence rate | Hurt |
| Decrease the variance of unlabeled inputs below 1 (ablation) | Iterative self-training · Synthetic Gaussian data | convergence rate | No clear effect |
| Increase the labeled-data weight parameter lambda within its feasible range (ablation) | Iterative self-training · Synthetic Gaussian data | relative weight error | Helped |
| Increase lambda beyond its feasible threshold (ablation) | Iterative self-training · Synthetic Gaussian data | relative weight error | Hurt |
| Add unlabeled images from 80 Million Tiny Images (ablation) | CIFAR-10 self-training · CIFAR-10 | test accuracy: up to 7% | Helped |
| Add unlabeled data (ablation) | Iterative self-training · Synthetic Gaussian data | labeled sample complexity for successful recovery | Helped |
Closest papers by meaning
- Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data (ICLR 2021 · similarity 0.93)
- Out-Of-Domain Unlabeled Data Improves Generalization (ICLR 2024 · similarity 0.86)
- Unlocking the Potential of Unlabeled Data in Semi-Supervised Domain Generalization (CVPR 2025 · similarity 0.86)
- On the Out-of-Distribution Generalization of Self-Supervised Learning (ICML 2025 · similarity 0.86)
- On the Importance of Embedding Norms in Self-Supervised Learning (ICML 2025 · similarity 0.85)
- Uncovering the Redundancy in Graph Self-supervised Learning Models (NeurIPS 2024 · similarity 0.85)
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.