Preprocessing for informative, efficient and small networks
Patrik Eklund, Lena Kallin Westin · 2002
We demonstrate how sigmoidal fuzzification affects discriminant capacities. In particular, we study the preprocessing perceptron and compare it with the multilayer perceptron. Case studies are selected from the medical domain, where output performance needs to be related to requirements for high sensitivities. These smaller and more informative networks tend also to be more robust with respect to accuracy with various requirements on sensitivities.