There are so many search methods in continuous optimization and discrete optimization, including population based PSO, GA, DE, EP, GP, ES, ACO, ABC,etc and single-point based search methods like SA, Tabu search, iterated guided search, Guided Local Search, Variable Neighborhood Search, and etc.
Sometimes it might be confusing about the classification in different literature.
The most general term would be metaheuristics and most of these methods are stochastic search/optimization methods. I doubt that there exist strict classifications of these methods. E.g., in evolutionary computation most of today's algorithms cannot be cleanly assigned to GA, ES, or EP, so it is better to call all of them EAs (the differences are more historic). Also, EAs and PSO have many common elements, but may use different names for similar concepts.
I agree with Helmut here as newer methods also apply combinations of these methods. Hence, I believe it probably does not make sense to strictly classify these methods.
All of them are metaheuristics but GA, ES,DE and GP are classified as evolutionary-based algorithms.
PSO, ACO, ABC, Cuckoo search , etc. are classifed as swarm intelligence-based algorithms. Tabu search is a local search algorithm
In my viewpoint, spirit of life (to solve the problem) is how to make the next generation better than our generation.
Humans do this by mimicking the strategy of birds, with genetic strategy ...., with the strategy of ants, ..... XX strategy, ... by the end of Day strategy.
What is important is that the strategy must be rational and reasonable, of course depending on the battle field.
So it is normal that sometimes all of these strategies are described in a way that is confusing.
Hi all,
I also agree that it is not easy and sometimes might not be necessary to classify the metaheuristic algorithms strictly.
I usually called those with mating selection or/and crossover as EA, but sometimes I also used a more specific name to emphasize the algorithm feature. I call it as GP when it uses tree-based representation, as ES when it uses only mutation, as DE when it uses difference between individuals in mutation, and as MA when it incorporates local search.
For trajectory-based search algorithms, I usually call it as SA when it accepts inferior solutions probabilistically, as TS when it prohibits visiting some solutions, as VNS when it uses multiple neighborhood functions, and as GRASP when it uses greedy strategies to construct solutions.
For swarm algorithms, I usually call it as PSO when it considers both group and individual observations and as ACO when it constructs solutions component by component based on memory.
Ok, that's a good question and surprisingly hard to answer. The most important difference is of course their names. However, they are very very similar (I would say identical). To explain, it is very easy to design a ES method just like PSO and call it a variant of ES. So, is PSO a ES or ES a PSO? Also, Newton method is an ES variant as well (a ES with deterministic mutation). So, basically even classical gradient methods can be counted as any method you like from EA community. Take a look at cs.adelaide.edu.au/~zbyszek/Papers/essay2.pdf, I have had a lot of discussions with the author and I agree with this essay. I hope it is helpful.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling