Benchmarking Real-Time Reinforcement Learning

Pierre Thodoroff; Wenyu Li; Neil D. Lawrence

Back to publications

Benchmarking Real-Time Reinforcement Learning

Pierre Thodoroff, Wenyu Li, Neil D. Lawrence

Pre-registration Workshop at NeurIPS 2021, 2021.

Abstract

Decision-making algorithms can require fast response time in applications as diverse as self-driving cars and minimizing load times of webpages. Yet, modern algorithms (deep reinforcement learning) are usually developed in scenarios where inference and training computational costs are ignored. This proposal aims to study reinforcement learning and control algorithms for real-time continuous control. In this scenario, the environment continuously evolves while actions are being computed by the agent (either in training or inference). The first goal is to provide a clear picture of the performance of modern algorithms modulated by their computational costs. The second goal is to identify the major challenges that arise when considering real-time environments to guide further research.

Links

Cite this Paper

BibTeX


@InProceedings{publications/benchmarking-real-time-reinforcement-learning,
  title = 	 {Benchmarking Real-Time Reinforcement Learning},
  author = 	 {Thodoroff, Pierre and Li, Wenyu and Lawrence, Neil D.},
  booktitle = 	 {Pre-registration Workshop at NeurIPS 2021},
  year = 	 {2021},
  pdf = 	 {https://preregister.science/papers_21neurips/26_paper.pdf},
  url = 	 {/publications/benchmarking-real-time-reinforcement-learning.html},
  abstract = 	 {Decision-making algorithms can require fast response time in applications as diverse as self-driving cars and minimizing load times of webpages. Yet, modern algorithms (deep reinforcement learning) are usually developed in scenarios where inference and training computational costs are ignored. This proposal aims to study reinforcement learning and control algorithms for real-time continuous control. In this scenario, the environment continuously evolves while actions are being computed by the agent (either in training or inference). The first goal is to provide a clear picture of the performance of modern algorithms modulated by their computational costs. The second goal is to identify the major challenges that arise when considering real-time environments to guide further research.}
}

Endnote

%0 Conference Paper
%T Benchmarking Real-Time Reinforcement Learning
%A Pierre Thodoroff
%A Wenyu Li
%A Neil D. Lawrence
%B Pre-registration Workshop at NeurIPS 2021
%D 2021	
%F publications/benchmarking-real-time-reinforcement-learning
%U /publications/benchmarking-real-time-reinforcement-learning.html
%X Decision-making algorithms can require fast response time in applications as diverse as self-driving cars and minimizing load times of webpages. Yet, modern algorithms (deep reinforcement learning) are usually developed in scenarios where inference and training computational costs are ignored. This proposal aims to study reinforcement learning and control algorithms for real-time continuous control. In this scenario, the environment continuously evolves while actions are being computed by the agent (either in training or inference). The first goal is to provide a clear picture of the performance of modern algorithms modulated by their computational costs. The second goal is to identify the major challenges that arise when considering real-time environments to guide further research.

RIS


TY  - CPAPER
TI  - Benchmarking Real-Time Reinforcement Learning
AU  - Pierre Thodoroff
AU  - Wenyu Li
AU  - Neil D. Lawrence
BT  - Pre-registration Workshop at NeurIPS 2021
DA  - 2021/12/13	
ID  - publications/benchmarking-real-time-reinforcement-learning
L1  - https://preregister.science/papers_21neurips/26_paper.pdf
UR  - /publications/benchmarking-real-time-reinforcement-learning.html
AB  - Decision-making algorithms can require fast response time in applications as diverse as self-driving cars and minimizing load times of webpages. Yet, modern algorithms (deep reinforcement learning) are usually developed in scenarios where inference and training computational costs are ignored. This proposal aims to study reinforcement learning and control algorithms for real-time continuous control. In this scenario, the environment continuously evolves while actions are being computed by the agent (either in training or inference). The first goal is to provide a clear picture of the performance of modern algorithms modulated by their computational costs. The second goal is to identify the major challenges that arise when considering real-time environments to guide further research.
ER  -

APA


Thodoroff, P., Li, W. & Lawrence, N.D.. (2021). Benchmarking Real-Time Reinforcement Learning. Pre-registration Workshop at NeurIPS 2021 Available from /publications/benchmarking-real-time-reinforcement-learning.html.