Changing Model Behavior at Test-Time Using Reinforcement Learning
arXiv:1702.07780
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.
Submitted to ICLR 2017 Workshop Track
Cited by in corpus (8)
- Dynamic Channel Pruning: Feature Boosting and Suppression
- Neural Speed Reading via Skim-RNN
- Dynamic Neural Networks: A Survey
- Dynamically Sacrificing Accuracy for Reduced Computation: Cascaded Inference Based on Softmax Confidence
- Surprisal-Triggered Conditional Computation with Neural Networks
- Dynamically Throttleable Neural Networks (TNN)
- Toward Runtime-Throttleable Neural Networks
- Fully Dynamic Inference with Deep Neural Networks