Research map / Learning theory and algorithms / Prediction and decision losses

Post-Processing Guarantees for Classification under Linear-Fractional Performance Metrics

NeurIPS 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Ranking-based loss optimization

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Characterizes the optimal binary classifier for linear-fractional metrics (F-score, Jaccard) as a thresholding of the regression function, reducing the problem to post-processing a conditional probability estimate.

Problem

Binary classification under generalized performance metrics beyond misclassification error, specifically linear-fractional measures such as the F-score and Jaccard index that matter in class-imbalanced settings.

Approach

Shows that the optimal classifier for linear-fractional metrics admits a thresholding structure on the regression function, so the problem reduces to choosing a threshold and post-processing a estimated regression function, with guarantees on the resulting classifier.

Closest papers by meaning

• Principled Algorithms for Optimizing Generalized Metrics in Binary Classification (ICML 2025 · similarity 0.89)
• Concentration and excess risk bounds for imbalanced classification with synthetic oversampling (NeurIPS 2025 · similarity 0.85)
• Leveraging Uncertainty Estimates To Improve Classifier Performance (ICLR 2024 · similarity 0.84)
• The Implicit Bias of Gradient Descent on Separable Multiclass Data (NeurIPS 2024 · similarity 0.84)
• A General Online Algorithm for Optimizing Complex Performance Metrics (ICML 2024 · similarity 0.84)
• Interplay of ROC and Precision-Recall AUCs: Theoretical Limits and Practical Implications in Binary Classification (ICML 2024 · similarity 0.84)