Multi-model Markov decision processes

Markov decision processes (MDPs) have found success in many application areas that involve sequential making under uncertainty, including the evaluation and design of treatment screening protocols for medical making. However, data used to parameterize model can influence what policies are recommended, multiple competing sources common areas, medicine. In this article, we introduce Multi-model process (MMDP) which generalizes a standard MDP by allowing models rewards transition probabilities. Solution MMDP generates single policy maximizes weighted performance over all models. This approach allows maker explicitly trade-off conflicting while generating same level complexity only consider source data. We study structural properties problem show it is at least NP-hard. develop exact methods fast approximation supported error bounds. Finally, illustrate effectiveness scalability our using case preventative blood pressure cholesterol management accounts published cardiovascular risk