RUBRIC: Realism--Utility Balanced Ranking for Imbalanced Classification

Name: RUBRIC: Realism--Utility Balanced Ranking for Imbalanced Classification
Uploaded: 2026-10-19T09:09:58.335420+00:00
Duration: 469 s
Description: RUBRIC introduces a generator-agnostic filtering framework to select synthetic data for imbalanced classification. It solves the problem of oversampling by optimizing a quality-over-quantity trade-off between sample realism and its utility in defining decision boundaries. This method tightens generalization bounds while boosting minority recall without sacrificing overall model performance.

RUBRIC introduces a generator-agnostic filtering framework to select synthetic data for imbalanced classification. It solves the problem of oversampling by optimizing a quality-over-quantity trade-off between sample realism and its utility in defining decision boundaries. This method tightens generalization bounds while boosting minority recall without sacrificing overall model performance.

Open the full episode