Controlled Deep Reinforcement Learning for Optimized Slice Placement

  • 2021-08-03 14:54:00
  • Jose Jurandir Alves Esteves, Amina Boubendir, Fabrice Guillemin, Pierre Sens
  • 2

Abstract

We present a hybrid ML-heuristic approach that we name "HeuristicallyAssisted Deep Reinforcement Learning (HA-DRL)" to solve the problem of NetworkSlice Placement Optimization. The proposed approach leverages recent works onDeep Reinforcement Learning (DRL) for slice placement and Virtual NetworkEmbedding (VNE) and uses a heuristic function to optimize the exploration ofthe action space by giving priority to reliable actions indicated by anefficient heuristic algorithm. The evaluation results show that the proposedHA-DRL algorithm can accelerate the learning of an efficient slice placementpolicy improving slice acceptance ratio when compared with state-of-the-artapproaches that are based only on reinforcement learning.

 

Quick Read (beta)

loading the full paper ...