Improving Model Classification by Optimizing the Training Dataset
This work addresses data efficiency for machine learning practitioners by improving classification performance through optimized coreset generation, though it appears incremental as it builds on existing coreset methods.
The paper tackles the problem of conventional coreset construction methods not optimizing for classification performance metrics like F1 score, and presents a systematic framework with tunable parameters that significantly outperforms both vanilla coresets and full dataset training on key classification metrics.
In the era of data-centric AI, the ability to curate high-quality training data is as crucial as model design. Coresets offer a principled approach to data reduction, enabling efficient learning on large datasets through importance sampling. However, conventional sensitivity-based coreset construction often falls short in optimizing for classification performance metrics, e.g., $F1$ score, focusing instead on loss approximation. In this work, we present a systematic framework for tuning the coreset generation process to enhance downstream classification quality. Our method introduces new tunable parameters--including deterministic sampling, class-wise allocation, and refinement via active sampling, beyond traditional sensitivity scores. Through extensive experiments on diverse datasets and classifiers, we demonstrate that tuned coresets can significantly outperform both vanilla coresets and full dataset training on key classification metrics, offering an effective path towards better and more efficient model training.