Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Hypothesis Testing by Betting on Predictions
Yaniv Tenzer, Elad Tolochinsky, Yaniv Romano
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 the imputed e-process (betting on predictions) to the convex combination of baselines | conv(LR Y, LR X, PPI) e-process · label shift simulation, low correlation, N=30 | power | Improved |
| add the imputed e-process to the baseline combination with abundant unlabeled data | conv(LR Y, LR X, PPI) e-process · label shift simulation, low correlation, N=135 | power | Improved |
| add the imputed e-process when correlation between X and Y is high and N=135 | baseline combination · label shift simulation, high correlation, N=135 | power | No clear effect |
| add the imputed e-process to the baseline combination under concept shift | conv(LR Y, LR Y|X, PPI) e-process · concept shift simulation, N=135 | power | Improved |
| add the imputed e-process to baselines for testing conditional improvement of one LLM over another | baseline combination · LLM evaluation, X = dataset identifier (GSM8K vs AQUA-RAT), n=10, N=90 | power | Improved |
| add the imputed e-process using LLM-judge annotations as X with few unlabeled samples | baseline combination · LLM evaluation, label shift, N=10 | power | Improved |
| add the imputed e-process using LLM-judge annotations with more unlabeled samples | baseline combination · LLM evaluation, label shift, N=60 | power | Mixed |
| use TabPFN in-context predictions in the imputed e-process with M=2 null datasets | conv(LR Y, PPI) e-process · California census private health insurance, 2017 vs 2019, n=50, N=100 | power | Improved |
| increase the number of sampled null datasets M (ablation) | imputed e-process · simulations (Appendix F.3) | power | Improved |
Closest papers by meaning
- Sequential Kernel Goodness-of-fit Testing (ICML 2024 · similarity 0.87)
- Protected Test-Time Adaptation via Online Entropy Matching: A Betting Approach (NeurIPS 2024 · similarity 0.86)
- Peeking with PEAK: Sequential, Nonparametric Composite Hypothesis Tests for Means of Multiple Data Streams (ICML 2024 · similarity 0.86)
- Optimal Algorithms for Augmented Testing of Discrete Distributions (NeurIPS 2024 · similarity 0.85)
- Prediction-Powered Active Testing (NeurIPS 2026 · similarity 0.85)
- Prediction-Powered E-Values (ICML 2025 · 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.
