Heuristic Optimization Methods Optimization as Search Chin-Shiuh Shieh
Optimization as Search Generate and Test initialize x do { generate x’ from x retain x x’ if F(x’) is better than F(x) } until terminate condition is met
Candidate Solutions x can be a single point in the solution space, or a population of candidate solutions.
Initialization Random initialization Incorporation of domain knowledge Re-initialization to cope with local optima
Generation of x’ from x “Direction” and “Step Size” Random vs. deterministic Random walk Gradient decent x’ x + Δd ; Large step size x’ x + δd ; Small step size x’ x + σu ; Random walk “Fusing” or “Exchange” explored knowledge, as in GA or PSO
Retain Best Solution Retain x’ if F(x’) is better than F(x). Sometimes retain x’ if F(x’) is worse than F(x) for the escaping of local optima.
Terminate Conditions No further improvement Bound of computational resource (time) is reached