Learning in real-world partially observable environments: Revisiting efficient deep Reinforcement Learning DVFS optimization for resource-constrained embedded devices

Moayadeldin Hussain

doi:10.1016/j.sysarc.2026.103797

Learning in real-world partially observable environments: Revisiting efficient deep Reinforcement Learning DVFS optimization for resource-constrained embedded devices

2026-04-04

0

PRE

AI

DOI：10.1016/j.sysarc.2026.103797

Original

Original request for help

Save

Abstract

En 中文

• Current DVFS solutions struggle on low-end devices. • Real environments complicate the modeling of RL-based solutions as fully observable MDPs. • We optimize the existing cutting-edge solution on Jetson Nano. • We reformulate the RL-based solution to partially-observable MDPs. • We consistently get significant improvements on 2 different tasks.

Keywords:

Deep Reinforcement Learning

DVFS Optimization

Resource-Constrained Devices

Partially Observable MDPs

Embedded Systems

Learning in real-world partially observable environments: Revisiting efficient deep Reinforcement Learning DVFS optimization for resource-constrained embedded devices

Abstract

Journal

Organization