Research map / Label-efficient and robust learning / Recognition and long-tail
Recovering the Propensity Score from Biased Positive Unlabeled Data
Walter Gerych, Thomas Hartvigsen, Luke Buquicchio, Emmanuel Agu, Elke Rundensteiner
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 |
|---|---|---|---|
| Estimate propensity scores using the Local Certainty approach | biased PU data with arbitrary propensity · PASCAL VOC and Yelp | propensity-score MAE: PASCAL VOC: 0.20+/-0.04 (Cluster) vs 0.13+/-0.03 (Local Certainty); Yelp: 0.21+/-0.07 (Cluster) vs 0.15+/-0.02 (Local Ce | Improved |
| Estimate propensity scores using the Probabilistic Gap approach | biased PU data with scaled propensity · PASCAL VOC and Yelp | propensity-score MAE: PASCAL VOC: 0.12+/-0.08 (Cluster) vs 0.09+/-0.05 (Probabilistic Gap); Yelp: 0.09+/-0.03 (Cluster) vs 0.04+/-0.02 (Probab | Improved |
| Use Local Certainty propensity estimates for downstream classification | biased PU classification with arbitrary propensity · HTRU 2, Adult, Bank, Wine, Yeast, and Digits | classification error: HTRU 2: 0.43+/-0.00 -> 0.04+/-0.00; Adult: 0.37+/-0.01 -> 0.35+/-0.02; Bank: 0.17+/-0.01 -> 0.06+/-0.01; Wine: 0.34+/-0. | Improved |
| Use Probabilistic Gap propensity estimates for downstream classification | biased PU classification with scaled propensity · Adult, Bank, Wine, Yeast, and Digits | classification error: Adult: 0.26+/-0.01 -> 0.25+/-0.03; Bank: 0.39+/-0.01 -> 0.30+/-0.01; Wine: 0.34+/-0.03 -> 0.25+/-0.02; Yeast: 0.44+/-0.0 | Improved |
| Use Probabilistic Gap propensity estimates instead of Cluster estimates | biased PU classification · HTRU 2, Scaled Propensity | classification error: 0.04+/-0.00 -> 0.05+/-0.00 | Hurt |
| Apply the Local Certainty estimator when class separability is not enforced | real-world PU data · PASCAL VOC and Yelp, Arbitrary Propensity | propensity-score MAE: PASCAL VOC: 0.13+/-0.03; Yelp: 0.15+/-0.02 | Improved |
| Use a label-indicator posterior model to estimate propensity | Local Certainty and Probabilistic Gap estimators · Additional experiments | propensity-score estimation | Hurt |
Closest papers by meaning
- PUe: Biased Positive-Unlabeled Learning Enhancement by Causal Inference (NeurIPS 2023 · similarity 0.90)
- Learning from Positive and Unlabeled Data with Arbitrary Positive Shift (NeurIPS 2020 · similarity 0.85)
- Positive-Unlabeled Learning With Label Distribution Alignment (TPAMI 2023 · similarity 0.85)
- PUATE: Efficient ATE Estimation from Treated (Positive) and Unlabeled Units (NeurIPS 2025 · similarity 0.85)
- Learning from positive and unlabeled examples -Finite size sample bounds (NeurIPS 2025 · similarity 0.85)
- Mitigating Spurious Correlations via Disagreement Probability (NeurIPS 2024 · similarity 0.82)
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.
