A nonextensive method for spectroscopic data analysis with artificial neural networks

Abstract

In this paper we apply an evolving stochastic method to construct simple and effective Artificial Neural Networks, based on the theory of Tsallis statistical mechanics. Our aim is to establish an automatic process for building a smaller network with high classification performance. We aim to assess the utility of the method based on statistical mechanics for the estimation of transparent coating material on security papers and cholesterol levels in blood samples. Our experimental study verifies that there are indeed improvements in the overall performance in terms of classification success and at the size of network compared to other efficient backpropagation learning methods.

Entropy- A Tale of Ice and Fire

“…Additional examples may be found in [16, 19, 24, 47, 48, 51, 52, 57-59, 79, 94, 97]. For the approximate value q = 1.3, see [3,14,17,20,23,30,32,33,53,55,61,73,126,129,133]. For the approximate value q = 1.4, see [7,13,17,21,22,25,78,83,115,128].…”

Convergence of Rprop and variants

“…The "globally convergent Rprop" (GRprop) is widely cited (e.g. [6,9,10,12,13,[15][16][17][18][19][20]), and is an option in at least one prominent software package for neural network modelling [21]. However, it has apparently gone unnoticed that the proof of GRprop's convergence is mistaken.…”