Evolutionary algorithms
When to use?
"Typical" optimisation
Multivariate functions, especially non-continous or hard to optimise with gradient-based methods.
Example: Rastrigin function.
Combinatorial optimisation
Problems with huge numerical complexity.
Example: given a list of cities find the shortest route through all of them.
In general
Evolutionary approach
A single iteration of a simple evolutionary algorithm:
Key characteristics
-
Algorithm is specific to each problem, no one-size-fits-all.
-
Evolutionary algorithms are inherently parallelizable on many levels.
