An optimization problem consists of finding, within a space ? of possible solutions S, the optimal solution S ^opt that minimizes (or maximizes) a given cost function F( S). From now on, we consider the minimization problem.

The classical deterministic optimization techniques that are devoted to solve this problem start with an initial guess S ₀, then this guess is iteratively improved until a given convergence criterion is satisfied. The output is an approximation S* of the optimal solution S ^opt. These techniques are based on the construction of a privileged search direction in ? that is determined using information on variations of F. Although they can be very effective, they usually converge to a local minimum in the vicinity of the starting guess S ₀. Hence, when the cost function F has numerous local minima, these techniques can easily miss the targeted global minimum. For such problems, stochastic optimization techniques are often preferred.

Stochastic optimization techniques are based on a random search that decreases the risk for the process to become stuck in a local minimum. They require only information on values of F, which makes their implementation easy. In what follows, we will first introduce some basic concepts and then we will focus on the Hill Climbing and the Simulated Annealing methods.