OPTIMIZING TENSOR PRODUCT COMPUTATIONS IN STOCHASTIC AUTOMATA NETWORKS

In this paper we consider some numerical issues in computing solutions to networks of stochastic automata (SAN). In particular our concern i s with keeping the amount of computation per iteration to a minimum, s ince iterative methods appear to be the most effective in determining numerical solutions. In a previous paper we presented complexity resul ts concerning the vector-descriptor multiplication phase of the analys is. In this paper our concern is with optimizations related to the imp lementation of this this algorithm. We also consider the possible bene fits of grouping automata in a SAN with many small automata, to create an equivalent SAN having a smaller number of larger automata. (C) Els evier, Paris.