Improving Data-Driven Reinforcement Learning in Wireless IoT Systems Using Domain Knowledge
Document Type
Article
Publication Date
11-1-2021
Abstract
Reinforcement learning (RL) algorithms are purely data-driven and do not leverage any domain knowledge about the nature of the available actions, the system's state transition dynamics, and its cost/reward function. This severely penalizes their ability to meet critical requirements of emerging wireless applications, due to the inefficiency with which these algorithms learn from their interactions with the environment. In this article, we describe how data-driven RL algorithms can be improved by systematically integrating basic system models into the learning process. Our proposed approach uses real-time data in conjunction with knowledge about the underlying communication system to achieve orders of magnitude improvement in key performance metrics, such as convergence speed and compute/memory complexity, relative to well-established RL benchmarks.
Identifier
85122423597 (Scopus)
Publication Title
IEEE Communications Magazine
External Full Text Location
https://doi.org/10.1109/MCOM.111.2000949
e-ISSN
15581896
ISSN
01636804
First Page
95
Last Page
101
Issue
11
Volume
59
Recommended Citation
Mastronarde, Nicholas; Sharma, Nikhilesh; and Chakareski, Jacob, "Improving Data-Driven Reinforcement Learning in Wireless IoT Systems Using Domain Knowledge" (2021). Faculty Publications. 3692.
https://digitalcommons.njit.edu/fac_pubs/3692