Swarm Intelligence - Science topic

IEEE XPLORE

Thank you Nikola Ivkovic for this information. Last year I met Prof. Milan Tuba at a conference and discussed PSO and its related applications. He is also concerned about a dedicated research lab on Swarm Intelligence. Still, the number of research labs is minimal in my opinion.

Yes, and there are lots of papers that have considered these kinds of problems. As for continuous state (or observation in POMDP), the DL network (whether DNN, LSTM, or CNN) can handle this. As for the second part of your question, there is no necessity that the action should affect the state space. If you only train your agent to properly act in different states, it is still an RL problem. There are several papers in wireless communications where the state (such as channels) is independent of the executed action while the action affects the reward (for instance, the data rate). However, it is important to properly design your state vector and add different features. Sometimes, it can be beneficial to integrate the previous action into the state vector and feed it to your DNN. It is noted that, if you work on a single-agent problem, usually the lower discount values are better when the state evolves independent of the action.

Dear Anita Rani Sulaf Waiss and

Yes, Python can be utilized for swarm optimization algorithms. Python provides a wide range of libraries and frameworks that support the implementation of swarm intelligence algorithms, making it a popular choice for swarm optimization.

Here are a few popular Python libraries for implementing swarm intelligence algorithms:

1. PySwarm: PySwarm is a Python library specifically designed for particle swarm optimization (PSO) algorithms. It provides an easy-to-use interface for defining optimization problems, configuring PSO parameters, and running swarm optimization simulations.

2. DEAP (Distributed Evolutionary Algorithms in Python): DEAP is a versatile evolutionary computation framework that includes support for various swarm intelligence algorithms. It provides a rich set of tools for implementing and evaluating swarm optimization algorithms, including genetic algorithms, particle swarm optimization, and ant colony optimization.

3. PyGMO (Python Parallel Global Multiobjective Optimizer): PyGMO is a Python library focused on multiobjective optimization. It includes several swarm optimization algorithms, such as particle swarm optimization, differential evolution, and artificial bee colony optimization.

4. SciPy: SciPy is a comprehensive scientific computing library in Python that provides functionality for optimization. It includes various optimization algorithms that can be adapted for swarm intelligence, such as differential evolution and simulated annealing.

5. Swarm Intelligence Algorithms in PyCairo: PyCairo is a Python library for implementing graphics and visualization. It can be utilized to visualize and analyze the behavior of swarm intelligence algorithms.

These libraries provide a convenient framework for implementing swarm optimization algorithms in Python. They offer tools for defining problem domains, initializing swarm populations, updating swarm positions, and evaluating fitness functions. Additionally, they often include functionalities for visualizing and analyzing the behavior and convergence of swarm algorithms.

When it comes to simulating swarm intelligence algorithms, Python-based tools and libraries can be used in a variety of software environments, including Jupyter notebooks, IDEs such as PyCharm or Spyder, or even command-line scripts. The choice of software depends on your specific needs and preferences.

Remember to refer to the documentation and examples provided by these libraries to understand their capabilities and to ensure proper utilization of the swarm intelligence algorithms in your simulations.

GitHub, GitLab, Bitbucket, and SourceForge are platforms for code sharing and collaboration. Search for Swarm intelligence, NS2, Reclassification algorithms, and Routing algorithms. In addition to these keywords, you may also consider searching for specific swarm algorithm keywords such as Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Genetic Algorithms (GA), and others. Combine these keywords with "NS2" to find repositories that implement these algorithms in NS2.

Before to choose an algorithm I suggest to read papers like

Camacho-Villalón, Christian L., Marco Dorigo, and Thomas Stützle. ‘Exposing the Grey Wolf, Moth-Flame, Whale, Firefly, Bat, and Antlion Algorithms: Six Misleading Optimization Techniques Inspired by Bestial Metaphors’. International Transactions in Operational Research

In this article a highly complex optimization problem such as optimal power flow is computed through various metaheuristics with varied performance.

Maybe I'm a little late but...

In PSO, min and max values usually will define the algorithm search space. I'm not that familiar with path planning, but , In your case it looks like the search space is the minimum and maximum values the path can assume. That is, if the path goes from point (0,0) to point (5,5), that would be the min and max values.

Metaheuristic algorithm and machine learning - MATLAB Central

https://www.mathworks.com ›

Engineering Optimization: An Introduction with Metaheuristic ... https://www.mathworks.com

metaheuristic-algorithms · GitHub Topics https://github.com › topics › metaheuristic-algorithms

Benghelima Slimane Charafeddine

This is a normal occurrence in all mathematical problems. There is no perfect solution that solves all problems, especially nonlinear problems. Always strive to improve and work harder. Thank you very much.

I would like to invite you to have a look on the following links.

SSO: https://www.mathworks.com/matlabcentral/fileexchange/92150-sperm-swarm-optimization-sso

HSSOGSA: https://www.mathworks.com/matlabcentral/fileexchange/92130-hssogsa?s_tid=srchtitle

MOSFP: https://github.com/sh7adeh1990/MOSFP

See the links:

The last few decades have seen the introduction of a large number of "novel" metaheuristics inspired by different natural and social phenomena. While metaphors have been useful inspirations, I believe this development has taken the field a step backwards, rather than forwards. When the metaphors are stripped away, are these algorithms different in their behaviour? Instead of more new methods, we need more critical evaluation of established methods to reveal their underlying mechanics.

Dear Robert Shour

This article might be useful, have a look

Kind Regards

Qamar Ul Islam

I recommend you to check also the CCSA algorithm implemented by a Conscious Neighborhood-based approach which is an effective mechanism to improve other metaheuristic algorithms as well. The CCSA and its full source code are available here:

Please read this article:

Medical Image Registration Using Evolutionary Computation: An Experimental Survey By: S. Damas; O. Cordón; J. Santamaría

By: Thomas Jansen

The 7^th section, namely "Results, Data Analysis, and Comparison", of the following current-state-of-the-art research paper have a sufficient answer for this question:

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.

Dear Chaomin Luo

I have read about the smart dust robot. It is still under investigation. You may go though , the project has a lot of limitations like ( power consumption, communication overhead, processing cost. Somehow you can get your aim.

regards

A hint: it is not OK to be ignorant of nonlinear mathematical optimisation. First you absolutely need to understand optimality criteria - as those conditions are the goal to fulfill. Mathematical optimisation tools are DESIGNED such that accumulation points from the iterative scheme WILL satisfy the optimality criteria, and in most cases you will be able to see the "fault" in the optimality criterion shrinking.

In contrast metaheuristics have no such "guiding light" at all - they fumble in the dark! Few sensible scholars with knowledge in mathematical optimisation would use these tools.

The increase of the population leads to (at least initially) increased diversity of the population, but if one increases the population size too much, there may be slow convergence of the population to the global optimum. But if the population is too small, it will lead to entrapment to local optima. It is widely suggested to increase the population in order to avoid local optima, especially if the objective function has many parameters. As for the number of iterations, an increased number will increase the possibility for convergence to global optimum, but you may start doing useless computations. In this case, you might need a good termination criterion that prevents useless computations after reaching global optimum.

For the latter, read the following paper.

Spanakis, C., Mathioudakis, E., Kampanis, N., Tsiknakis, M., & Marias, K. (2016). A Proposed Method for Improving Rigid Registration Robustness. International Journal of Computer Science and Information Security, 14(5), 1.

Dear Long Nguyen Cong,

Can you able to generate Levy random number in the interval [a, b] using Levy distribution.

Hi

It is usually convenient for an optimization algorithm to obtain the optimal solution when the optimal point is in the symmetric search space. When the optimal point of a test function is shifted, the search space goes out of symmetry. Therefore, the power of the algorithm in detecting optimal solutions may be reduced.

Dear D\ Ahmed

No general methods for dimensionality reductions because of the various of data characteristics. To overcome this limitation, you can use ensemble learning (as, ensemble feature selection). It supports the diversity and stability terms. I recommend you this paper " Ensembles for feature selection: A review and future trends":

https://www.sciencedirect.com/science/article/pii/S1566253518303440

This book for more details "Recent Advances in Ensembles for Feature Selection",

https://link.springer.com/book/10.1007%2F978-3-319-90080-3

Also check this paper " Swarm Intelligence Algorithms for Feature Selection: A Review

"

I think Its really depend on how you initialize the population.

I assume that you randomize population S with n solution, then take the opposition population S^ based on S. Now, there are two possible ways you can do here is:

+ 1st: Compare each solution in S and S^ with their respective index. Take the solution with better fitness.

+ 2nd: You concatenate S and S^ then take n solution with the best fitness.

Both ways have the weakness is: Maybe, all position of all solution placed at the 1 side of search space. Especially, when that side belongs local optima, It will make your algorithm fast convergence (but in local optima)

I think the best way to use the opposition-based technique for the initial population is:

You randomize population S with n/2 solution, then create n/2 solution left with opposition-based technique. This way make sure that all position doesn't fall in the 1 side of the search space.

All the answers are great

I have already proposed several improved versions. This book Will be published soon:

Dear Anik,

I suggest you to see links and attached files on topic.

Paper Title (use style: paper title) - laccei

Survey on Unmanned Aerial Vehicle Networks: A Cyber ... - arXiv

Survey of Important Issues in UAV Communications Networks - arXiv

Best regards

I think switching topology case can be considered as a subset of time-varying formation/consensus. Consider the following two cases for discrete time consensus:

(i) When the topology of a network is switching, the system matrix W[k] (row-stochastic for discrete-time) will be different for some k>0 such that w_ij[k] > 0 does not mean w_ij[k'] > 0 for some k' ≠ k.

(ii) Suppose that topology is fixed but agents of a network use different weights for some k that results in time-varying system matrix W[k]. In that case, if w_ij[k] > 0 holds, then we have w_ij[k'] > 0 for some k' ≠ k.

Similar arguments are also valid for Laplacian matrix L(t) (zero row sum for cont. time). Hence, while (i) represents the switching topology case, time-varying formation/consensus case might be represented either (i) or (ii) depending on the problem.

Hope this helps.

You surely use some heuristic to find an approximate Pareto front. You will not able to find the Pareto front. To approach better to it, in the supposition that you generate a solutions population, you should to simulate the real behavior of your process, for each solution in the population to recalculate the the multiple objective function value and to reorder the found solutions. You won't reach exactly the Pareto front but you will approach enough and to obtain very good close to the Pareto front solutions

There is already some papers on it.

One example:

At the moment, it does not sound very promising but I think eventually these two fields will be married.

But again, we have to find a proper application for this marriage. All I have seen until now is some made up application which does not sound convincing. Maybe your Masters Thesis will change that.

thanks for your ans.

Hi Eduard,

My first recommendation is to choose a proper platform and programming language suitable for Concurrency , distributed computing and scalability , easy to implement NN. Erlang is what I am using at the moment, I found it is perfect for this type of implementation; any other platform would cost you enormous amount of investment to get anywehere.

Then you need an architecture from bottom-up. scalability is very important in the solution your seeking.

See this:

I'm not able to sent code for you, but this should be quite easy to re-implementation.

Dear Prof Amal

I think this link may be of your help.

Neutrosophic logic application

With best regards

Hi Leonardo,

I understand that you might be busy.

Please find time for answer.

Waiting for favourable response.

Thanks in advance.

1.- Computational cost: SI-based meta-heuristics are relatively easy methods to implement.

Normally this type of algorithms performs searches in continuous spaces, and two approaches can be used to use them in discrete spaces: 1) Modify the operators that update the particles in the swarm, or 2) to define the scheme to map a candidate solution of the problem in a particle of the swarm.

These approaches impact the performance of an SI-based method.

2.- Performance metrics: This depends on what you want to measure: time, space, precision, for example. It also depends on the problem that is being solved.

Dear Samira Esheghi,

of course i am agree with our colleague Luy Tan Nguyen and I add that exist many papers about this topic using differents techniques. For more proof i suggest you to see links and attached files in this topics.

-Cooperative Control of Multi-Agent Systems - Optimal and | Frank L ...

- Distributed Cooperative Control of Multi-agent Systems

https://books.google.dz/books?isbn=1119246202 -

- Cooperative Control of Multi-Agent Systems: Theory and Applications

https://books.google.dz/books?isbn=1119266122 -

- Cooperative Control of Multi-Agent Systems: Theory and Applications ...

https://www.wiley.com/.../Cooperative+Control+of+Multi+Agent+...

- Cooperative Control of Distributed Multi-Agent Systems

https://books.google.dz/books?isbn=0470724196 -

Best regards

I have developed algorithm and railway traffic management system inspired by bee foraging technique and decision theory. maybe you can find out more from the papers that i have published

Hi, I recently published a paper entitled " Swarm robotics search & rescue: A novel artificial intelligence-inspired optimization approach", so-called SRSR optimization algorithm. I also shared MATLAB code of this algorithm as well as 19 other well-known heuristic evolutionary algorithms in form of one free package for the sake of a fair comparison on the Internet. Based on my experiences, each algorithm can be suitable for a specific set of problems based on its inherent features and formulation; in terms of convergence quality and convergence speed. If an algorithm has outstanding performance on some problems does not necessarily guaranty that can be outstanding for other problems as well. Many people believe that some algorithms like PSO are completely dated and you should not apply them to any problem anymore; However I found that traditional PSO has a better performance than several new algorithms in some special case-studies. Finally, there is no other option in my idea, you should test a bunch of algorithms to find best one suited to your problem. Moreover, some of algorithms have a good performance in small-scale problems and others are better in case of large scale. In my vision, quality of an algorithm is 100% percent related to the problem you wanna to apply on. Some time, some algorithms suffer from low pace because their computational cost is high, however they can find very high quality solutions. On the other side, they may have high convergence speed with lower convergence quality.

From another point of view, publication of more than tens evolutionary algorithms during a year, as you said, may assist other researchers to find some novel more applicable ideas in case of applied soft computing; because this is a step-by-step evolutionary procedure. Some time the best option is to simply combine to algorithms enjoy abilities of both ones and solve your problem; however it is not a new optimization algorithm.

Kind regards

Mojtaba

What a real problem do you mean?

yes you can do that , i advise you to perform two tests also the first one is t test and other one is w test. very easy and efective to measure and execute 

Hello,

Samer Sarsam,

Thank you so much to have pointed out this.Yes its true, PSO is one of the methods that can be used to optimize SVM. Infact, similar work has been done by me wherein i have used SVM as a base classifier and thereby incorporated ACO, PSO and Flower Pollination (Swarm based methods) to optimize the base classifier SVM.

SVM does not optimizes PSO, and I have not mentioned that, however SVM can be used to control some of the parameters of PSO or ACO or FPAB. As i have mentioned in the last line 

The answer i has written earlier was in context to: For instance, SVM are capable of delivering high performance in terms of classification accuracy but their proficiency truly relies on an optimal choice of hyper-parameters. On the other hand, Control parameters selection of PSO has no theoretical guidance, most choices of PSO are based on experience. 

That is wat was intended..of course SVM is not an Optimization technique, it is just a learning model that can be used as a base classifier. Thank you for sharing the article.

Regards,

Rebika Rai

I am working on finding the optimal path using ant colony.

if u can use PSO then u can use QPSO too but for me Quantum is used to enhance the pso method but the ability of quantum is as heuristic as PSO. so it depends on what kind of result u r depicting from yo algorithm.

How a colony of O. smaragdina distributes its workforce and territory iIs a key behavior in the biological pest management using this ant.  There are two processes (1) nest-allocation to each tree (how many nests are constructed in a specific tree), and (2) worker-allocation to each food items (how many workers take part in when they bring food item) and each part of the territory.

      a. Our preliminary study suggests a strange phenomenon that we observed in each oil palm tree whereby the ant constructs the odd number of nests. We examine if this phenomenon does really exist (not only on oil palms but on other trees).

      b. If the above phenomenon exists we experimentally study its mechanism by manipulating the number of nests in the field.

      a. Examine the relationship between the size of the food item and group size by video especially focusing on its movements.

      b. Monitor the daily activity of foragers in the field to know when they hunt what items on what time?

A national survey is conducted on the total number of nest occupying each palm trees in Peninsular, Sabah & Sarawak, related to a unique phenomenon of oddness number of nests. The survey is carried out during both rainy and dry season of the year.

Direct counting is done using a binocular device for tall palms.  This is a novel study on an exceptional behavior rarely seen in nature. The phenomena of odd prime number exist only among the insect periodical cicadas Magicicada septendecim, M. septendecula, and M. cassini in North America with a life cycle of emergence after 13 or 17 years (Fontaine et al., 2007). Little is known on this unique factor related to the odd number of 13 and 17 (it is never 12 or 18; even numbers). A recent report related the cicadas timing of emergence to the abundance and synchrony of avian predators populations in the USA. Cuckoos (Coccyzus spp.) increased highly during emergence years to decline consequently profusely while red-bellied Woodpeckers (Melanerpes Carolinus), Blue Jays (Cyanocitta cristata), Common Grackles (Quiscalus quiscula) and Brown-headed Cowbirds (Molothrusater) had their population abundance increase for 1-3 years subsequently to cicadas emergence to finally decline (Koenig & Liebhold, 2005). What are the underlying biological mechanisms responsible for this phenomenon? in the case of O. smaragdina nesting behaviors in oil palm plantations.

To investigate the underlying biological mechanism, to understand this, performing the following experiments might lead to verify the existing nesting behavior among Oecophylla smaragdina ants devising the tendency of constructing an odd prime number of nests per palm tree.  From the previous survey of occupancy:

I-                   By adding a nest to a tree that already has a nest, and observe what happen (if the populations of the added nest are always absorbed to the old one?).

II-                Or release a few collected but disturbed nests on the base of a palm tree where already a nest exists on the canopy. 

III-             From these, the expected hypothesis is - Ants whose nests were collected and disturbed may climb up to the nearest palm tree to reconstruct a new nest.  However, on the tree, there is already a nest belonging to the same polydomous colony.  If there is any mechanism that regulating the nest numbers leading to the odd prime number phenomenon, ants should never reconstruct a nest, but if they do they will make 2 or 4 new nests in both cases

IV-             . Another trial is to destructively cut a frond of a palm tree having i.e 11 nests to artificially reduce the total nest number to 10. In that case, if there is a real apparatus and variable responsible or explaining this mysteriousness, the ants shall rebuild at least one new nest to rebalance the tendency of odd nest number to obtain again 11 nests in total. It is possible that three more nests would be added for a final total of 13 nests.

The trial is statistically analysed by Chi Square test X2, and data are verified by Fisher’s exact test or Barnard’s exact test which is a more powerful alternative of the previous for samples size n < 1000. Barnard’s exact test can verify the strength of the data record on odd prime total nest number per palm tree obtained.

I 

Thanks a lot!

Normally we normalize input variable to reduce the chances of getting stuck in local optima. It also makes training MLP faster  When inputs are normalized in range of [0 1] use logsig() activation function and when in range of [-1 1] use tansig() activation function instead  In your case (function approximation), you can use linear activation function (purelin) activation function for output neuron. For other applications (e.g., classification, clustring) you better use sigmoidal (tansig or logsig) activation functions for all layers (hidden layer(s) as well as output layer). Good luck.

Don't hesitate to contact me if you need any help. My main research topic is continuous black box optimization, within the TAO team ( same team of authors of the BBOB/COCO testbench )

Dear Muhammad Gulraj,

As Dr. Ramon López de Mántaras pointed out Bee or Ant colonies foraging is a behavior that is simulated to solve optimization problems. Grey Wolf Optimizer(GWO) algorithm mimics the leadership hierarchy and hunting mechanism of grey wolves in nature.  Particle Swarm Optimization (PSO) simulates the social behavior of bird flocking or fish schooling which is used to solve different optimization problems. For more information please see the following links:

Regards,

Dear Moodys,

Here are link and attached files in subject.

-A Hybrid CS/PSO Algorithm for Global Optimization - Springer

Best regards

Hi Rafid,

I work with AI methods. They excellently apply on function generation and data prognosys. I am not familiar with any application based on this methods for Enterprise Business Application testing. My suggestion is to make one for youreself, based on AI methods (Fuzzy, c-slusters, neuron network, ANFIS).

good luck

The local search can be very destructive or very constructive. It depends on the type of problems that is attempted to be solved.  It is worth exploring whether or not the chaos can bring something positive to the search or not. The "no-free-lunch" theorem suggests that one techniques may be optimum with one problem, but its performance may be degraded for another. So it is worth to bear this in mind. 

Hi Nikolaos,

In my experience, normalization procedure is used in NN input if the different features that are being used are in a different numeric value/ranges. The normalization is done feature-wise usually, so the impact on the network would be similar.

The sigmoid function is not applied directly to the input features, I mean, if we have X = [x_1, ... , x_n] as input an W_i = [w_i1, ..., w_in ] the synaptic weights corresponding to the i'th neuron of the following hidden layer, the activation of that neuron w.r.t. the sigmoid function would be:

Y=sigmoid(X · transpose(W_i)) that it's always between the range (0,1).

If you start with very large values in the input and weights, the sigmoid signal would be 'crushed' to 1.

There are several ways to initialize the initial weights, but small random numbers is usually fine. At the end, these weights will be updated according to an optimization function to complete the designed task.

Hope that some of this information helps you.

Cheers.

There are many examples of quality measures in the literature. Some aim at measuring the distance of an approximation set to the Pareto-optimal front: Van Veldhuizen [21], e.g., calculated for each solution In the approximation set under consideration the Euclidean distance to the closest Pareto-optimal objective vector and then took the average over all of these distances. 

Other measures try to capture the diversity of an approximation set, e.g., the chi-square-like deviation measure used by Srinivas and Deb [18]. A Further example is the hypervolume measure which Considers the volume of the objective space dominated by an approximation set [26]. In these three cases, an approximation set is assigned a real number which is meant to reflect (certain aspects of) the quality of an approximation set. Alternatively, one can assign numbers to pairs of approximation sets.Zitzler and Thiele [26], e.g., introduced the coverage function which gives for a pair (A, B) of approximation sets the fraction of solutions in B that are weakly dominated by one or more solutions in A.

Zitzler, E., Thiele, L., Laumanns, M., Fonseca, C. M., & Da Fonseca, V. G. (2003). Performance assessment of multiobjective optimizers: an analysis and review. Evolutionary Computation, IEEE Transactions on, 7(2), 117-132.

The simplest way to represent a particle is a vector. For example, in an optimization problem with 3 design variables, each design (particle) is represented by a vector [x1, x2, x3]. So each particle is such a vector and represents a point in the 3D space (that we all know and can imagine). For problems with dimension > 3 it is difficult to imagine the particle and its position, but with N=3 you get the idea.

Also, please read the following publication which has helpful information about PSO and its implementation in structural engineering problems:

Dear Please follow below links. Hope it will help

Dear @Sidartha, I find this article is very relevant for your issue.

"Dr. Louis Rosenberg, CEO of Unanimous AI, is building a software platform, UNU, that assembles groups to make collective decisions. "What's different about this is that it fundamentally keeps people in there," he said. "We're focused on using software to amplify human intelligence."...

UNU is an online platform where anyone can log in and answer questions as a swarm. "People benefit when they form a real-time system the way bees, fish, and birds do, as opposed to people just casting a single vote the way people do," said Rosenberg..."

Maybe some resources are available through this article.

Hi, I am also working on this problem using PSO. In fact, number of particles/ants or number of iterations depends upon your problem. You can experiment by giving different values of particles and by setting different number of iterations and check results. Initially, you may set small number of iterations and small number of particles.

You should balance between number of particles and number of iterations.