There is a lot of applications in bioinformatics.

I'm currently working with metaheuristics applied to multiple sequence alignment. We can see many approaches based on genetic algorithm, ant colony optimization, among others.

Regards.

As long as you also derive lower bounds (in order to assess the optimality gap), and compare (where it is possible) to a software that will generate an optimal solution, I'm all for it - otherwise most probably not.

Article Metaheuristics -- the metaphor exposed

This is already some years old, and already then the shit had hit the fan.

I plan to apply, ... ... or I have already done so in (research unpublished so far). It will be useful to continue discussing possible options in this regard. For example, the question of "swarm intelligence" which I previously perceived as a kind of cliché. But my recent research has revealed that there are some additional features that provide intelligent ( highly promising, advanced) solutions. It would be intriguing to share some relevant literature on the subject.

Metaheuristic is a boon to the optimization society. The possibilities are endless. I am currently working on enhancing heat transfer rate through fins by optimizing important variables using swarm intelligence.

Chapter How Meta-heuristic Algorithms Contribute to Deep Learning in...

Article Deep Reinforcement Learning for Solving the Vehicle Routing Problem

Thank you Michael for sharing Metaheuristics -- the metaphor exposed; it says the true if there is no real applications for so many introduced metaheuristic algorithms. Also thank Denis and all of you, but I really still looking for their applicability in the real applications. Do you know any projects and applications especially related to data and big data using these algorithms?

Regards

Dear Dr. Mohammad H. Nadimi-Shahraki

May I ask you expain what is your purpose of "projects and applications"?

As you know, many feature selection algorithms in Data Mining use the Metas. If you are searching for an industrial application like optimizing a/some parameter(s) or threshold(s), as others mentioned, there are many application for Metas there too. For example task scheduling in cloud computing usually use Meta algorithm, not only in research papers, either in real worlds.

So, if you had a special purpose, please clarify.

Sincerely

Your Student

I have used metaheuristic schemes for a set of applications such as Image segmentation, shape detection, block matching and corner detection.

Being Scandinavian I have read and seen countless of times some representation of the children's story about the emperor's new clothes - by Hans Christian Andersen.

As you might know the emperor's new costume is soooo fine that it is weightless, and no-one - but a child - dares to speak up to the emperor and exclaim that he in fact is naked! Thanks to the child, the curtain opens, and we truly see that he indeed is naked.

I find metaheuristics - to some degree - to be that, too, within the mathematical optimization domain: quite shallow, devoid of solid theory, and often (but not always) a game of draw, guess, and jump (I do not know the exact English translation of the name of the children's game). I have no truck with it when we are dealing with very complicated combinatorial/integer/bilevel problems in industry, especially when we do not have an explicit formulation, such as when we need to deal with the use of simulations within the optimization, or uncertain coefficients. But then we are talking about industrial mathematics, which is something else than mathematics - which is an exact science.

The theme of metaheuristics was, I hope, originally an attempt to find "reasonable" solutions to the most difficult and large problems, especially for nonlinear integer models in industry, with the explicit sign that with these techniques we might hope get a fairly good solution if we are lucky, or we may not - as that is actually how it works: metaheuristics are NOT globally convergent in general, and they were - make a mental note of this - NOT EVEN CONSTRUCTED TO BE.

Yet there are plenty of scholars - especially in this forum, for a reason I do not fully understand - that insist on applying their favourite metaheuristic(s) on just anything. Yesterday I think it was when I found a paper on a metaheuristic used to solve a very, very simple linear program with one (1) linear constraint. I blew my top, as they say. WTF is going on? I blew my top because I know how to solve to guaranteed optimality such a problem in under a 1/1000 of a second on a slow computer.

Can any sane person closer to that field address this, please? It is irrational, to begin with; or it is simply the fact that the world of scientific endeavours no longer are defined by codes of conduct? I am really troubled by more and more often seeing this unscientific methodology being used, and I sure hope it never will be seriously compared with mathematical optimization. Well, in fact, it is compared every day in industry, and math always wins.

https://www.datasciencecentral.com/profiles/blogs/swarm-optimization-goodbye-gradients

More info can be found here:

N. K. T. El-Omari, "Sea Lion Optimization Algorithm for Solving the Maximum Flow Problem", International Journal of Computer Science and Network Security (IJCSNS), e-ISSN: 1738-7906, DOI: 10.22937/IJCSNS.2020.20.08.5, 20(8):30-68, 2020.

Or simply refer to the same paper at the following address:

https://www.researchgate.net/publication/344324283_Sea_Lion_Optimization_Algorithm_for_Solving_the_Maximum_Flow_Problem

Indeed - and max-flow solvers can solve huge instances faster than your finger can let go of the RETURN key!

So why on Earth does this still go on, when every one should have found out that metaheuristics only should be utilised when we have messy problems that we do not what math prog tool to use? And perhaps even then we shouldn't.

You're going to be quite sad when you hear that a competitor has used a mathematical optimisation model based on solving the max-flow problem optimally, thrown away all heuristics, and perhaps has also implemented the solver that will yield profit to both parties.

And the sea lion will be looking for a clue, and never find one.

Preprint Combining reinforcement learning with lin-kernighan-helsgaun...

To tackle different properties of real optimization problems, a new multi-trial vector approach named MTV was proposed in the following link. MTV combines different search strategies in the form of trial vector producers (TVPs). In the MTV approach, the TVPs are applied on their dedicated subpopulations, which are distributed by a winner-based distribution policy, and share their experiences efficiently by using a lifetime archive. By using the MTV approach, you can design your own TVPs and develop an effective algorithm to solve a real optimization problem such as optimal power flow (OPF) and truss optimization problems.

https://www.sciencedirect.com/science/article/abs/pii/S1568494620306992

Amir: You are dishonest when your title states "Optimal design ...", when you only run a meta-heuristic.

It is also dishonest when the result from a meta-heuristic never (as far I have seen) is compared to a run of a mathematical optimisation method that guarantees to find an optimum.

I am not interested in teasing you because we get the optimum and you probably not, but to have a frank discussion on the pros and cons of hard problems, and how to reach feasible solutions that are acceptable. I have read several articles where the theme is a complex problem, which is "attacked" by a metaheuristic, and I find it puzzling that the articles lack discussions on how to get even better solutions, or ways to combine mathematical optimisation tools and (meta)heuristics - one evident idea is to run a few heuristics and from there apply a mathematical optimisation tool. I think that they ought to be rather good in many applications.

The maximum flow problem is one of the easiest problems to solve to optimality. It adds nothing to the literature when you compare different metaheuristics on a very simple problem like that. It is only a contribution if your metaheuristic can solve an easy problem like the maximum flow or the shortest path problem much faster than a conventional optimization algorithm that is designed for it. However, it is not likely that metaheuristics will get there, because you have to make much larger number of computations and comparisons even for a very simple problem, compared to optimization algorithms. Metaheuristics should be applied to harder problems, problems that cannot be efficiently solved by optimization algorithms; and if they produce good results within reasonable time, then we can see some contribution there.

Here, here - at last a friend! :-). Thanks, Cenk! I have had several discussions with metaheuristics enthusiasts, and have pointed out what is not fruitful. One might hope to discuss what problems to attack with what tools - as it is the combination of problem AND algorithm to use that is essential.

Article Machine Learning at the service of Meta-heuristics for solvi...

I am interested in geophysics. As you know, metaheuristics indeed can solve various problems, inversion problems in geophysics are not the exceptional. The essence of inversion problems in geophysics are optimal parameters obtaining (which is very suitable for metaheuristics to solve), with the well-defined parameters, geophysical models can yield more proper results, hence, maybe these problems can trigger more sparks for you.

Metaheuristic procedures are not designed to terminate at an optimal solution, which can be seen clearly in how the procedures are defined. It's no use talking about whether they terminate far away or not - the fact is that there is no guarantee about the output at all, and that fact should make the switch from them immediate. But no - there are very many who insist on accepting inferior output. I find that both strange and sad - and you are a loser because of it.

Well, one field where metaheuristics have gained both momentum and traction is in structural design. Due to the highly non-linear design constraints and cumbersome objective functions associated with them, no known analytical techniques have yet been devised to address this problem. Ergo, algorithms have been both conceived and implemented to obtain better designs and are applied to real world problems. Here's one of my papers where truss structures are optimized with metahueristics: https://www.sciencedirect.com/science/article/abs/pii/S2352012421006421

Personally, I am more interested in the application of the metaheuristics in solving real-world problems, the ones that mathematical optimization methods are hitherto unable to solve, rather than trying to devise what is called NOVEL metaheuristics. However, one might want to enhance the performance of already-existing ones for their especial purposes such as structural optimization problems. For instance, I used a metaheuristic and, by some simple yet efficient techniques, improved its performance for optimal design of some structures with complex configurations such as barrel vaults and grids. You can check out the article here:Article An Improved Water Strider Algorithm for Optimal Design of Sk...

An example of application of a swarm intelligence metaheuristic algorithm in the Arctic ship design: https://www.researchgate.net/publication/351618850_A_Holistic_Multi-Objective_Design_Optimization_Approach_for_Arctic_Offshore_Supply_Vessels

Check this particular application in Cloud Computing..

Same can be applied to any cluster or network for QoS optimization.

https://www.mdpi.com/2076-3417/11/14/6244/htm

As an another application, meta heuristic algorithms could be employed in training of ANN especially when we have some unnatural loss functions or regularization terms. The superiority of this approach over conventional training methods such as back propagation has been proved in several papers. I can refer to two of my publications entailing Article An ANN-based ensemble model for change point estimation in c...

Article Using evolutionary artificial neural networks in monitoring ...

Please do not hesitate for more questions.

Best regards.

Dear Professor Michael Patriksson, I read your comments and found them insightful. I wonder if you suggest any alternatives to use as optimization tools for problems that are complex, include many local optimums and lack derivative information such as structural optimization problems.

It can be used to solve complex scheduling problems, space allocation problems, and clustering problems.

The truth is that many metaheuristics that have been proposed recently seem to be less relevant than they claim. I share with you some references where the subject is discussed.

- Aranha, C. et al. (2021) ‘Metaphor-based metaheuristics, a call for action: the elephant in the room’, Swarm Intelligence, (0123456789). doi: 10.1007/s11721-021-00202-9.

- Tzanetos, A. and Dounias, G. (2021) ‘Nature inspired optimization algorithms or simply variations of metaheuristics?’, Artificial Intelligence Review, 54(3), pp. 1841–1862. doi: 10.1007/s10462-020-09893-8.

In my humble opinion, the important thing in this field is to understand the principles that allow a metaheuristic to find high quality solutions, focus on the theoretical foundations rather than on the most recent metaheuristics. That will allow you to concentrate your time on relevant research papers on optimization instead of wasting it trying to identify what is the innovative part or the original contribution provided by papers presenting "new" metaheuristics.

You can check out this linkArticle Applications of metaheuristics in real-life problems

https://www.hindawi.com/journals/mpe/2017/8306732/

Article Application of Heuristic and Metaheuristic Algorithms in Sol...

Recently we published an article on the application of metaheuristic algorithms in the Optimal Power flow problem, considered a problem in power engineering. You can check the below link

Article Metaheuristic Algorithms in Optimal Power Flow Analysis: A Q...