Improving Model Classification by Optimizing the Training Dataset

TL;DR

This paper introduces a systematic framework for tuning coreset construction to improve classification performance, demonstrating significant gains over traditional methods and full dataset training.

Contribution

It presents a novel, tunable coreset generation approach that optimizes for classification metrics like F1 score, beyond traditional sensitivity-based methods.

Findings

Tuned coresets outperform vanilla coresets in classification tasks.

The method improves F1 scores significantly across datasets.

Coreset-based training reduces data size while maintaining accuracy.

Abstract

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.