Difference between arbitrary Evolutionary and Genetic Algorithm

Question

I'm new to this site. I am not really sure if this the right section for this question - it might belong to computer science.

Recently I started working on my master thesis about solving multi-objective optimization problems using an evolutionary algorithm with a limited number of decison maker calls. I want to motivate why this is a reasonable approach and therefore explain how an evolutionary algorithm works. But here I found myself confronted with a problem: Most of the time all the authors talk about evolutionary algorithms but than they start talking abut genetic algorithms. I do understand that genetic algorithms are a sub-class of evolutionary algorithms but I do not understand how they differ from other evolutionary algorithms. Somethimes I get the feeling both terms are even used equivalently. Could someone please explain to me if and how they are different from each other (main properties or working principles of both classes)?

I know this is a broad question but I hope it still fits the terms of this site.

Answer 1

As I understand it (and others may disagree), there are several differences between Genetic Algorithms (GA) and Evolutionary Strategies (ES), as traditionally conceived:

a) Traditionally, design points in GA are encoded as bit-strings [however there ARE alternatives for some problems], and the operators which act on them manipulate these strings; so if the design points represent sets of Real parameters, these are not varied in a very 'numeric' manner. Standard ES usually operate directly on the Real parameters as such.

b) A primary operator in GA is crossover, which mimics biological reproduction by combining two (sometimes more) design points by applying a 'cut-and-splice' to the 'chromosomes' (i.e the bit-strings) [and yes, there is a theoretical explanation of why this makes sense]. ES usually only 'mutates' design points, e.g by varying one of the Real parameters at a time.

c) It has been my observation that GA will generate an entirely new population of design points each round, using the 'crossover' and 'mutation' operators. ES, on the other hand, seems to only modify one or a few designs at a time.