85Chapter four: Systems optimization techniques
A main problem in the solution of these problems is the numerical compu-
tation and differentiation of risk functions.
Stochastic approximation
While some nonclassical optimization techniques are able to optimize
on discontinuous objective functions, they are unable to do so when
complexity of the data becomes very large. In this case, the complex-
ity of the system requires that the objective function be estimated.
Furthermore, the models that are used to estimate the objective function
may be stochastic due to the dynamic and random nature of the system
and processes.
The basic idea behind the stochastic approximation method is the gra-
dient descent method. Here the decision ...