STRIPAI: Determining the Suitability of Implementing Deep Reinforcement Learning Principles into new domains
Summary
Deep Learning and Reinforcement Learning are techniques that are being applied more and more. However, the combination of the two techniques sees little use outside of the gaming domain. This study aims to determine when and how these techniques can be used in new domains. This thesis asks the question how the two techniques can be applied to applications in new domains in order to improve the usability of these applications.
Based on the used methodology, a literature research on how to determine whether a domain is actually suitable for the application of Deep Learning and Reinforcement Learning is conducted. After this, a framework is shown which can be used to transform an application in such a way that it represents a game with rules, inputs, and output. Lastly, the techniques will be applied to an existing application in the prescriptive healthcare domain.
The results indicate that applying these techniques to the application in the prescriptive healthcare domain did not lead to a significant increase in the effectiveness of the application. Furthermore, theoretical results showed that there was also no significant increase in the efficiency of the application. This means that the implementation of Deep Learning and Reinforcement Learning principles did not lead to a significant increase in application usability. The results from this experiment can be used in order to better determine which domains will be suitable for these kinds of implementations.