“…Many of the neural network training algorithms, such as the simultaneous perturbation stochastic approximation algorithm (Spall, 1992), the Widrow Hoff algorithm (also known as the "least mean square" algorithm) (Haykin, 1999, pp.128-135), the Alopex algorithm (Harth & Tzanakou, 1974) and self-organizing maps (Kohonen, 1990), can be regarded as special instances of stochastic approximation. Refer to Bharath & Borkar (1999) for more discussions on this issue. Recently, stochastic approximation has been used with Markov chain Monte Carlo for solving maximum likelihood estimation problems (Gu & Kong, 1998;Delyon et al, 1999).…”