Machine learning models are often used at test-time subject to constraints and trade-offs not present at training-time. For example, a computer vision model operating on an embedded device may need to perform real-time inference, or a translation model operating on a cell phone may wish to bound its average compute time in order to be power-efficient. In this work we describe a mixture-of-experts model and show how to change its test-time resource-usage on a per-input basis using reinforcement learning. We test our method on a small MNIST-based example.
@article{arxiv.1702.07780,
title = {Changing Model Behavior at Test-Time Using Reinforcement Learning},
author = {Augustus Odena and Dieterich Lawson and Christopher Olah},
journal= {arXiv preprint arXiv:1702.07780},
year = {2017}
}