Neural Networks and the Estimation of Hands' Strength in Contract Bridge

This paper is focused on a Double Dummy Bridge Problem (DDBP) which consists in answering the question of how many tricks are to be taken by a pair of players assuming perfect play of all four sides with all cards being revealed. Several experiments are also presented in a variant of DDBP in which the information about to whom of the two players in a given pair a particular card belongs to is hidden. In contrast to our previous works, which were devoted to no trump contracts, here we concentrate on suit contracts. Several interesting conclusions are drawn from comparison of weight patterns of networks trained on no trump contracts only vs. those trained exclusively on suit contracts. The ultimate goal of this research is to construct a computer program playing the game of contract bridge using neural networks and other CI techniques with the basic assumption of using zero-human-knowledge approach and to learn purely on examples.