Data-Driven Deep Reinforcement Learning Framework for Large-Scale Service Composition

抄録

In this research, reinforcement learning is used to select service component for SOA. QoS is the evaluation criterion of service component and it is used to represent payoff. Considering real application links to the problem that the number of interaction with environment is limited in real application. Offline RL, which learns their policy function from fixed interaction data, is one of method to solve this. There was little work to focus on application of RL to SOA in the offline setting. In this research, We focus on application RL to the setting where the part of service component is changed. Offline RL enables learning using a smaller number of data than conventional online methods, and that pre-learning of models can be performed even when the environment changes.