Information enhancement—A tool for approximate representation of optimal strategies from influence diagrams Original Research Article

The main source of complexity problems for large influence diagrams is that the last decisions have intractably large spaces of past information. Usually, it is not a problem when you reach the last decisions; but when calculating optimal policies for the first decisions, you have to consider all possible future information scenarios. This is the curse of knowing that you shall not forget. The usual approach for addressing this problem is to reduce the information through assuming that you do forget something (Nilsson and Lauritzen, 2000, LIMID ), or to abstract the information through introducing new nodes (Jensen, 2008) . This paper takes the opposite approach, namely to assume that you know more in the future than you actually will. We call the approach information enhancement. It consists in reducing the space of future information scenarios by adding information links. We present a systematic way of determining fruitful information links to add.