Understanding of Evolutionary Algorithms in Artificial Intelligence (AI) 2019: Where and How to Apply?

Analyzation of different Evolutionary algorithms in Artificial intelligence (AI)

September 05, 2019

Human behaviour is simulated using an artificial intelligence (AI) technique. AI provides a better, quicker and more accurate alternative to an optimisation problem than existing conventional techniques. In general, an artificial intelligence technique uses multiple approaches to achieve an ideal solution. In this paper, three models based on evolutionary algorithms (EAs) are introduced, analysed and compared. These EAs include the genetic algorithm (GA), the particle swarm optimisation (PSO), and the ant colony optimisation (ACO).

Genetic Algorithm

A genetic algorithm is a heuristic search influenced by the theory of natural evolution proposed by Charles Darwin Yang, (2013). This algorithm represents the natural selection process in which the most suitable individuals are chosen for reproduction to generate next-generation children. Genetic algorithms (GA) are neural networks which are influenced biologically and constitute a fresh computational model based on evolutionary sciences (Meyer-Baese & Schmid, 2014).

In a genetic algorithm, five stages are considered

1. Initial population
2. Fitness function
3. Selection
4. Crossover
5. Mutation

The working process of the genetic algorithm is,

• Population is initiated
• Individual fitness is evaluated
• Individual rank is evaluated
• The new population, selection and crossover were generated.
• The process repeated until it finds the optimal solution.

Particle Swarm Optimization (PSO) Algorithm

Particle swarm optimization (PSO) was developed by Kennedy and Eberhart. The author was inspired by the social organism’s group behaviour like ant colonies, bird flocking and fish schooling (Martínez & Cao, 2019). Particle swarm optimisation depends on a bird group; here in the PSO algorithm, the group of birds is called as ‘swarm’. This algorithm imitates information sharing between members. The evolutionary computation techniques, the PSO has many similarities with Genetic Algorithms (GA). Using the population of random solutions, the system is initialized and optima searched via generation update. Unlike GA, however, PSO does not have any evolutionary operators like mutation and crossover.

The working process of the PSO algorithm is,

• Randomly initialized, All particle positions vectors(Solutions)
• Using the simulation model, the objective function and the associated constraint is evaluated.
• In each particle best vector position is determined
• Particles best global vector position is determined
• The position and velocity of each particle were calculated and updated.
• The process repeated until it finds the optimal solution.

Ant Colony Optimization (ACO) Algorithms

Ant colony optimisation (ACO) algorithm is based on the ant’s ability to find the shortest way to a food source from the nest (Deb, 2011). An ant constantly hops from one place to another and finally achieve the target (food) (Christensen & Bastien, 2016).

The ACO algorithm consists of the following steps:

• The concentration of pheromones is initially determined.
• Randomly initialized, all ant position vectors (i.e., solutions)
• Using the simulation model, the objective function and the associated constraint are evaluated for each ant position.
• The concentration of pheromones corresponding to each path (variable values) is updated.

Summary of the comparison result of GA, PSO and ACO

Indicator	GA	PSO	ACO
Accuracy	Provides a feasible solution near to optimal. Wihartiko, Wijayanti, & Virgantari  (2018)	Provides optimal solution (Wihartiko et al., 2018)	Superior in finding the optimal solution  Selvarajan, Samath, Jabar, and Ahmed, (2019)
Iteration	GA needs more variable and it has more constraint, so more iteration is required.	PSO also needs more variable and constraints. However, it takes less time than GA	ACO needs more variable and constraint, so more iteration is required.
Additional Method required	In certain generations, it takes the addition of chromosome extermination techniques in order to obtain the optimal solution	No additional technique is required	No additional technique is required
Encoding Time	Reduction of encoding time is less when compare to PSO and ACO	Efficiently reduce the encoding time	Efficiently reduce the encoding time
Image quality	The image quality is less when compared to PSO and ACO	Same quality image is maintained	Same quality image is maintained
Compression Ratio	High	Very High	Very High
Computation	Efficient	Efficient	Efficient

Conclusion:

The three evolutionary algorithms such as the genetic algorithm (GA), the particle swarm optimisation (PSO), and the ant colony optimisation (ACO) have been analysed and compared in Table 1. All three algorithm is excellent in the performance and but GA has few constraints. when compared with PSO and ACO, the GA is complicated in execution and it also consumes time to execute. Furthermore, the GA requires more iteration, so it requires more variable compared to the PSO and SCO. In the future, we can use GA for complex processes and effective results. Where else, we can use the PSO and ACO algorithm for small and simple project execution.

References