Research map / Label-efficient and robust learning / Test-time adaptation
Coupled Training with Privileged Information and Unlabeled Data
Jiahao Shi, Omar Hagrass, Jason Matthew Klusowski
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 |
|---|---|---|---|
| couple deployment and rich-view models in joint training instead of Two-Stage pseudo-labeling | Coupled Training (random forest) · Parkinson's Telemonitoring | test MSE: 83.32 -> 72.60 | Improved |
| Two-Stage pseudo-labeling with privileged features versus X-only supervised baseline (ablation) | Two-Stage · Parkinson's Telemonitoring | test MSE: 77.34 -> 83.32 | Hurt |
| couple deployment and rich-view models instead of X-only baseline | Coupled Training (linear) · Bank Marketing, 26 seeds | Brier score: 0.0893 -> 0.0881 | Improved |
| Two-Stage pseudo-labeling with corrupted privileged features versus X-only baseline (ablation) | Two-Stage · Bank Marketing, 26 seeds | Brier score: 0.0893 -> 0.0928 | Hurt |
| use squared-loss generalized distillation instead of Two-Stage (ablation) | Two-Stage · Bank Marketing, 26 seeds | Brier score: 0.0928 -> 0.0907 | Improved |
| use SVM+ (labeled-only privileged information) instead of X-only baseline (ablation) | SVM+ · Bank Marketing, 26 seeds | Brier score: 0.0893 -> 0.1470 | Hurt |
| increase coupling weight lambda from small to intermediate (ablation) | Coupled Training · Parkinson's Telemonitoring | test MSE | Mixed |
| large lambda (approaches Two-Stage) versus intermediate lambda (ablation) | Coupled Training (logistic) · synthetic classification | test error | Hurt |
| add small coupling when privileged signal is weak (aggressive pseudo-responses, small nu) (ablation) | Coupled/Two-Stage regime · synthetic linear Gaussian, weak privileged signal | estimation error | Hurt |
| increase number of unlabeled samples (ablation) | Coupled Training · synthetic linear controls | test error | Improved |
| increase nuisance privileged dimensions (ablation) | Coupled Training vs Two-Stage · synthetic linear controls | test error | Mixed |
| use AFS-based dictionary selection with intermediate lambda | Coupled Training with AFS · PneumoniaMNIST (n=100 labeled, 2000 unlabeled) | test AUROC | Improved |
Closest papers by meaning
- Self-Joint Supervised Learning (ICLR 2022 · similarity 0.86)
- On the Provable Advantage of Unsupervised Pretraining (ICLR 2024 · similarity 0.84)
- Unextractable Protocol Models: Collaborative Training and Inference without Weight Materialization (NeurIPS 2025 · similarity 0.84)
- Specialized Foundation Models Struggle to Beat Supervised Baselines (ICLR 2025 · similarity 0.84)
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.84)
- Learning with Synthetic Data via SGD in High-Dimensional Linear Regression (NeurIPS 2026 · 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.
