A self-organizing neural network for the FE simulation of light propagation

A new approach, for developing a finite element grid for the computation of light transfer in virtual three-dimensional environments is presented. In this work the finite element, mesh, is instantiated by the internal structures of two artificial neural networks which are trained for learning the three-dimensional geometry and the light flow relationships. Based on the networks´ topologies the finite element method is performed. Through the use of a neural network type. with specific outstanding function approximation, and clustering facilities, the approach outperforms classical approaches in its accuracy and in its robustness concerning huge geometries where classical approaches run into a lack of computing resources