Research map / Label-efficient and robust learning / Recognition and long-tail
Fisher Consistency for Prior Probability Shift
Dirk Tasche
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 |
|---|---|---|---|
| use ACC or EM instead of CDE-Iterate under prior probability shift | CDE-Iterate · Binormal test, population, Q[Y=0]=0.1 | estimated class 0 prevalence: 0.0121 -> 0.1000 | Improved |
| run CDE-Iterate to the second iteration instead of the first (ablation) | CDE-Iterate · Binormal test, population, Q[Y=0]=0.1 | estimated class 0 prevalence: 0.2269 -> 0.1131 | Improved |
| run CDE-Iterate to convergence instead of the second iteration (ablation) | CDE-Iterate · Binormal test, population, Q[Y=0]=0.1 | estimated class 0 prevalence: 0.1131 -> 0.0121 | Hurt |
| replace ACC by EM under invariant density ratio shift | ACC · Non-normal test with binormal density ratio, population, Q[Y=0]=0.1 | estimated class 0 prevalence: 0.0958 -> 0.1000 | Improved |
| use CDE2 instead of ACC outside prior probability shift | ACC · Non-normal test with binormal density ratio, sample, Q[Y=0]=0.1 | estimated class 0 prevalence: 0.0828 -> 0.1014 | No clear effect |
| use EM when density ratio is not invariant | EM · Non-normal test with non-binormal density ratio, population, Q[Y=0]=0.01 | estimated class 0 prevalence: 0.0100 -> 0.1307 | Hurt |
| use ACC when test set deviates from prior probability shift | ACC · Non-normal test with non-binormal density ratio, population, Q[Y=0]=0.99 | estimated class 0 prevalence: 0.9900 -> 0.8230 | Hurt |
| use CDE-infinity instead of CDE1 or CDE2 when density ratio is square-rooted | CDE-Iterate · Non-normal test with non-binormal density ratio, population, Q[Y=0]=0.01 | estimated class 0 prevalence: 0.2664 -> 0.0000 | Improved |
Closest papers by meaning
- On the Consistency of Ordinal Regression Methods (JMLR 2017 · similarity 0.85)
- Bayes Consistency vs. H-Consistency: The Interplay between Surrogate Loss Functions and the Scoring Function Class (NeurIPS 2020 · similarity 0.85)
- Estimating Accuracy from Unlabeled Data: A Bayesian Approach (ICML 2016 · similarity 0.85)
- On the Characterization of a Class of Fisher-Consistent Loss Functions and its Application to Boosting (JMLR 2016 · similarity 0.84)
- Leveraging Self-Consistency for Data-Efficient Amortized Bayesian Inference (ICML 2024 · similarity 0.83)
- Bernstein–von Mises for Adaptively Collected Data (NeurIPS 2025 · 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.
