This paper presents Explorer, a neural‐network‐based learning system, which combines Artificial Intelligence techniques with classical numerical methods to classify the dynamical behavior of a set of experiments or simulations. Explorer is easy to use and to modify, and allows one to perform qualitative studies of dynamical models at a low cost. This new tool can be adapted to many different situations because it does not require any prior theoretical background about the model being studied. Explorer first stores the relevant knowledge under the user’s supervision. Then it is used in stand‐alone mode to perform an automatic exploration of a predefined domain of parameters. Explorer is used to study a complex nonlinear dynamical model, the Beasts’ model, associated with a class of physical problems known as ‘‘percolative transport.’’ The exploration map obtained allows one to visualize the distribution of the behavioral regimes in the chosen parameter space, and to point out the role played by each parameter: One parameter leads the system to chaos as it increases, while the other acts to bring the system back to a converging trajectory.
Skip Nav Destination
Research Article| November 01 1992
Learning dynamical behaviors with explorer
Bruno Fromont, Pascale Beaufumé; Learning dynamical behaviors with explorer. Comput. Phys. 1 November 1992; 6 (6): 660–669. https://doi.org/10.1063/1.168423
Download citation file: