Linh Hoang

Linh Hoang

Data Scientist @ Rakuten

PORTFOLIO

Deep Reinforcement Learning for Network Resource Management

View on Github Open on IEEE Xplore Open on IEEE Xplore

Designed an actor-critic DRL framework to assign mobile users to edge servers based on their traffic and network conditions. The objective is to minimize energy consumption given a delay threshold and with queue stability constraints. The actor module is based on a CNN (using Tensorflow), and the critic module is backed by Lyapunov optimization. The method reduces the energy to a 3% gap compared to an exhaustive search. The research was partly presented at IEEE ICC 2022 (won the IEEE ComSoc Travel Grant) and published in IEEE/ACM Transactions on Networking.