When problem knowledge is indicated by a couple of past data, a neural network can be utilized and trained with the data. The network should then be treated as a connectionist information base. The data used can be exact or fuzzy or a combination of both.