QR-SDN: Towards Reinforcement Learning States, Actions, and Rewards for Direct Flow Routing in Software-Defined Networks.
Justus RischkePeter SossallaHani SalahFrank H. P. FitzekMartin ReissleinPublished in: IEEE Access (2020)
Keyphrases
- reinforcement learning
- perceptual aliasing
- state action
- reward function
- markov decision processes
- partially observable
- action selection
- network topologies
- state transitions
- action space
- initial state
- dynamic routing
- learning algorithm
- fully observable
- state space
- software development
- function approximation
- partial observability
- reinforcement learning algorithms
- state and action spaces
- optimal policy
- network structure
- software systems
- dynamic programming
- wireless ad hoc networks
- shortest path
- temporal difference
- traffic engineering
- path selection
- transition model
- social networks
- markov decision problems
- machine learning
- partially observable domains
- model free
- state variables
- network topology
- software components
- network design
- reward signal
- packet forwarding
- behavioural cloning
- differentiated services
- state abstraction
- network nodes
- state information
- belief state
- singular value decomposition