Biometric Authentication: A Machine Learning Approach

5.4. The Back-Propagation Algorithms

The most prominent algorithm for training multi-layer networks is the so-called back-propagation (BP) algorithm. The algorithm, independently proposed by Wer-bos [377], Parker [269], and Rumelhart et al. [326], offers an efficient computational speedup for training multi-layer networks. The objective is to train the weights w_ji(l) so as to minimize the mean-squared error E in Eq. 5.3.1. The basic gradient-type learning formula is

Equation 5.4.1

with the n-th training pattern, a⁽ⁿ⁾(0), and its corresponding teacher t⁽ⁿ⁾, n = 1, 2,. . ., N, presented. The derivation of the BP algorithm follows a chain-rule technique: ...