Neuroevolution, or neuro-evolution, is a type of artificial intelligence that generates artificial neural networks (ANN), parameters, and rules using evolutionary algorithms. 

It is utilised most frequently in artificial life, general game playing, and evolutionary robotics. Neuroevolution's primary advantage is that it can be deployed more broadly than supervised learning methods, which require a curriculum of accurate input-output pairings. In contrast, neuroevolution requires measuring a network's task performance. For instance, the outcome of a game (i.e., whether a player won or lost) can be measured without identifying a preferred strategy. Neuroevolution is typically employed as part of the reinforcement learning paradigm. It can be contrasted with standard deep learning approaches that use gradient descent on a neural network with a fixed topology.

Differences

Numerous neuroevolution algorithms have been defined. One prominent contrast is between algorithms that develop only the strength of the link weights given a fixed network topology (often referred to as conventional neuroevolution) and those that evolve both the network topology and its weights (called TWEANNs, for Topology and Weight Evolving Artificial Neural Network algorithms).

Methods that evolve the structure of ANNs in parallel with its parameters (those using traditional evolutionary algorithms) can be distinguished from those that develop them separately (through memetic algorithms).

With ANN

It is an artificial neural network-based AI function, a bio-inspired approach to computational intelligence and machine learning. Neuroevolution, on the other hand, is a branch of artificial intelligence and machine learning that use evolutionary algorithms to build artificial neural networks.

With gradient-descent

Neuroevolution is not used as often as gradient descent in neural networks. But around 2017, researchers at Uber said they had found that simple structural neuroevolution algorithms could compete with modern gradient-descent deep learning algorithms, which are the industry standard. 

Journalist Matthew Hutson wrote in Science that he thinks the increase in computing power in the 2010s is one reason why neuroevolution is working now when it didn't work before. Furthermore, there is a link between neuroevolution and gradient descent, which can be shown.

Examples

Neuroevolution is most commonly used in reinforcement learning, evolutionary robotics, and artificial life. Examples include evolving behaviours for board and video games, directing mobile robots, and studying the evolution of biologically relevant behaviours.

Advantages

As with all evolutionary algorithms, the benefit of neuroevolution is that a population of ANNs can be quickly and naturally processed in parallel. So, for example, if there are 100 ANNs in the people and 100 processors, you can evaluate all those networks simultaneously in the time it takes to assess one network.

Applications

Neuroevolution has been successfully implemented in numerous domains, such as 

• strategy games, 
• image processing and computer vision, 
• text mining and natural language processing, 
• speech processing, 
• software engineering, 
• time series analysis, 
• cybersecurity, 
• finance and fraud detection, 
• social networks, 
• recommender systems, etc.