Adaptive Data Selection for Text Classification: Hierarchical Value with a Local Utility Budget
The Adaptive Data Selection for Text Classification method addresses selecting optimal data subsets for training neural networks under a fixed budget. It proposes Data Value Contribution (DVC), which evaluates data quality and relevance across six metrics—layer-wise and global—to guide selection. This budget-aware approach outperforms existing methods, showing significant accuracy gains while maintaining sublinear computational scaling.