Near-optimal sample compression for nearest neighbors
Machine Learning
2018-03-28 v4 Computational Complexity
Abstract
We present the first sample compression algorithm for nearest neighbors with non-trivial performance guarantees. We complement these guarantees by demonstrating almost matching hardness lower bounds, which show that our bound is nearly optimal. Our result yields new insight into margin-based nearest neighbor classification in metric spaces and allows us to significantly sharpen and simplify existing bounds. Some encouraging empirical results are also presented.
Cite
@article{arxiv.1404.3368,
title = {Near-optimal sample compression for nearest neighbors},
author = {Lee-Ad Gottlieb and Aryeh Kontorovich and Pinhas Nisnevitch},
journal= {arXiv preprint arXiv:1404.3368},
year = {2018}
}