Value of Information Analysis in Dynamic Decision Models

Decision-making under dynamic environments is an important problem frequently encountered in the practice. In this paper, we reveal some properties of value of information in dynamic environment, and compare the results with that in different representations by other researchers. We study the time invariant features in dynamic influence diagrams (DID) to enhance the value of information calculation in dynamic decision models. We identify the structure in DIDs which could be decomposed to temporal invariant sub-DIDs and then be constructed into sub-junction trees to improve the efficiency in calculation of the total expected value and the value of information. We also discuss issues including finding spreading variables which will complicate the calculation, adding mapping variables directly into the sub-junction trees to enable a simple calculation for value of information for decisionintervening variables, and discounting benefits among the sub-junction trees according to time. Finally, we propose an algorithm based on these discussions which is polynomial to the size of the network when there are no spreading variables in the DID.