
128
Chapter 7
twice the maximum anticipated distance between the starting guess
and the best point.
Random Number Generation
Random numbers play an important role in neural network learning.
In the simplest cases, they are used to initialize weights in preparation
for learning. More sophisticated programs use them in simulated
annealing and genetic optimization, discussed in the next chapter.
Some neural network models even use random numbers as a central
ingredient in their operation, rather than only in training. For these
reasons, and because simulated annealing is especially dependent on
good random numbers, they will be briefly discusse ...