What is an Evolutionary Algorithm?

An Evolutionary Algorithm is based on the ideas from biological evolution and its developmental pattern. It is an optimisation procedure that can find new approaches and solutions to specific problems. Evolutionary algorithms are also called genetic algorithms.

Such a method is ideally suited for more complex problems. In this way, whole individuals can be created in a simulation and these receive a so-called fitness function with fitness values. This fitness value provides information about how well such an individual can solve the corresponding problem. Afterwards, selection is carried out, whereby the best adapted individuals "survive".

In the whole process, two important elements are considered prerequisites for improvement, namely recombination and mutation. The strongest properties are secured by the process and the AI approaches the optimal solution. Changes become smaller and smaller and may even occur as soon as the maximum is reached. This process can be manually interrupted beforehand to avoid the problem of over-adaptation.

What is the difference between Evolutionary Algorithms and Backpropagation?

Evolutionary algorithms are an alternative to Backpropagation. In backpropagation, a gradient method is used to continuously improve the results. Error analysis and root cause analysis are used. Afterwards, further and further adjustments are made. Evolutionary algorithms behave much more "randomly" when the weighting parameters are changed.

Backpropagation is used in the context of supervised learning. Evolutionary algorithms, on the other hand, can also be used in unsupervised learning.The results are used in pattern recognition and reinforcement learning.

Where do evolutionary algorithms come into play?

Evolutionary algorithms can be used beyond optimisation and search. They are used in art, modelling and simulation. Other fields of application are the development of sensor networks, stock market analysis and the prediction of RNA structures.

In combinatorics, an optimal solution cannot always be derived due to the high computational effort. Heuristics are often used for such optimisations. Important for the use of evolutionary algorithms are the fitness function, mutation and variation, with which generations of new elements can be calculated.