Test Sample Accuracy Scales with Training Sample Density in Neural Networks
Abstract
Intuitively, one would expect accuracy of a trained neural network's prediction on test samples to correlate with how densely the samples are surrounded by seen training samples in representation space. We find that a bound on empirical training error smoothed across linear activation regions scales inversely with training sample density in representation space. Empirically, we verify this bound is a strong predictor of the inaccuracy of the network's prediction on test samples. For unseen test sets, including those with out-of-distribution samples, ranking test samples by their local region's error bound and discarding samples with the highest bounds raises prediction accuracy by up to 20% in absolute terms for image classification datasets, on average over thresholds.
Cite
@article{arxiv.2106.08365,
title = {Test Sample Accuracy Scales with Training Sample Density in Neural Networks},
author = {Xu Ji and Razvan Pascanu and Devon Hjelm and Balaji Lakshminarayanan and Andrea Vedaldi},
journal= {arXiv preprint arXiv:2106.08365},
year = {2022}
}
Comments
CoLLAs 2022 oral