Anytime Tail Averaging
Machine Learning
2019-02-21 v2 Applications
Machine Learning
Abstract
Tail averaging consists in averaging the last examples in a stream. Common techniques either have a memory requirement which grows with the number of samples to average, are not available at every timestep or do not accomodate growing windows. We propose two techniques with a low constant memory cost that perform tail averaging with access to the average at every time step. We also show how one can improve the accuracy of that average at the cost of increased memory consumption.
Cite
@article{arxiv.1902.05083,
title = {Anytime Tail Averaging},
author = {Nicolas Le Roux},
journal= {arXiv preprint arXiv:1902.05083},
year = {2019}
}
Comments
Added a specific section on the case of multiple accumulators when k_t is a constant