Quality-Based Reinforcement Learning in Intelligent Opportunistic Software Composition
Kahina Hacid, Sylvie Trouilhet, Françoise Adreit, Jean-Paul Arcangeli · 2021
Internet of Things and cyber-physical systems are characterised by openness and an increasing number of devices and their associated services. In a previous work, we have proposed to exploit opportunistically these services in order to automatically make emerge customised applications that suit user preferences. For that, we have developed a generic solution for bottom-up opportunistic service composition, based on reinforcement learning. In this work, it is extended to handle more efficiently the appearance of new components using service annotation and quality attributes in order to generalise and share knowledge with new discovered services. A didactic use case is used for illustration and demonstration purposes.