COMPARATIVE DISCUSSION ON GLOBAL MODELS FOR OPTICAL PROPERTIES OF Bi2O3

The novelty of this paper is due to the investigation of the possibility of reaching a reliable global effective model of temperature dependence on the optical properties, based on automated neural network (ANN). For this approach, starting with a series of experimental data set of the measured reflectance magnitudes we built and studied 20 different ANN with RBF and MLP structure and with different activation functions of neurons. After the training, validation and test stages, only the best 10 models were retained and a systematic comparison between the simulations results is included in the paper. The discussion is continued with the comparison of ANN approach results obtained for refractive index magnitudes. As in the previous part, after the training, validation and test stages, only the best 10 models were retained and a systematic comparison between the simulations results. The obtained results are promising but the performance is not higher than in the previous analytical investigation.