Anytime Tail Averaging
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.
- Publication:
-
arXiv e-prints
- Pub Date:
- February 2019
- DOI:
- 10.48550/arXiv.1902.05083
- arXiv:
- arXiv:1902.05083
- Bibcode:
- 2019arXiv190205083L
- Keywords:
-
- Computer Science - Machine Learning;
- Statistics - Applications;
- Statistics - Machine Learning
- E-Print:
- Added a specific section on the case of multiple accumulators when k_t is a constant