Learning internal representations by error propagation
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Probability Density Estimation Using Adaptive Activation Function Neurons
Neural Processing Letters
| Hi-index | 0.01 |
The minimisation of a least mean squares cost functionproduces poorresults in the ranges of the input variable where the quantity to beapproximatedtakes on relatively low values. This can be a problem if an accurateapproximationis required in a wide dynamic range. The present paper approachesthis problemin the case of multilayer perceptrons trained to approximate theposteriorconditional probabilities in a multicategory classification problem.The use of acost function derived from the Kullback–Leibler information distancemeasure isproposed and a computationally light algorithm is derived for itsminimisation.The effectiveness of the procedure is experimentally verified.