A neural net architecture for various learning processes on AC skin analysis

In previous work by Chang and Charleson (1994), the direct measurement of AC skin parameters based on a neural net architecture is faster than conventional methods because it uses a real-time, on-line measurement and computation, and requires only one high frequency input signal. In general, deciding whether to use the Widrow learning law or its variants for any problem in neural nets depends on the individual case. In this paper, the author presents the various learnings in a neural net architecture for AC skin parameters estimation. After computer simulations, the results show that the accuracy of estimates for any learning methods are almost no different. However, the time-consuming of using the Widrow learning is about three times as long as that of using its variant learning. As a result, the variant learning has a significant effect on the performance of AC skin parameters estimation to speed the convergence.