Comparative analysis of importance sampling techniques to estimate error functions for training neural networks

The application of importance sampling to train neural networks which approximates the Neyman-Pearson detector is considered in this paper. A comparative study with two different error functions is carried out. These two error functions are selected to make the Neyman-Pearson detector approximation possible. The importance sampling technique is used to estimate the error function for training. Some results are presented to compare the performance of both approaches to approximate the optimum detector. Furthermore, results show the convenience of using the importance sampling technique for training neural networks, when low probabilities of false alarm are considered