Changing Model Behavior at Test-Time Using Reinforcement Learning
Abstract
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.
Cite
Text
Odena et al. "Changing Model Behavior at Test-Time Using Reinforcement Learning." International Conference on Learning Representations, 2017.Markdown
[Odena et al. "Changing Model Behavior at Test-Time Using Reinforcement Learning." International Conference on Learning Representations, 2017.](https://mlanthology.org/iclr/2017/odena2017iclr-changing/)BibTeX
@inproceedings{odena2017iclr-changing,
title = {{Changing Model Behavior at Test-Time Using Reinforcement Learning}},
author = {Odena, Augustus and Lawson, Dieterich and Olah, Christopher},
booktitle = {International Conference on Learning Representations},
year = {2017},
url = {https://mlanthology.org/iclr/2017/odena2017iclr-changing/}
}