Artificial Neural Networks in Chemistry

In this project theoretical aspects and practical applications of artificial neural networks are explored. In the theoretical part the emphasize is given to the problem of 2D maps of neural network weights and their interpretations. On the basis of this research we have shown that the arguments saying that, although quite good for making predictions as a nonlinear modelling tool, the ANNs are useless in the interpretation of the results they deliver. The close study of 2D maps of internal counter-propagation ANN weights has shown that logical rules in the form (''''if ... then'''') can be derived by overlaping these weight maps. From the application point of view various type of problems have been solved: from QSAR modelling of biologcal activities to clustering of archeological findings.