Neuroevolution: Fundamentals and Applications for Surpassing Human Intelligence with Neuroevolution

By Fouad Sabry

What Is Neuroevolution

Neuroevolution, sometimes spelled neuro-evolution, is a form of artificial intelligence that generates artificial neural networks (ANN), parameters, and rules through the application of evolutionary algorithms. Neuroevolution is also spelled neuro-evolution. The most popular applications for this technique are found in evolutionary robotics, artificial life, and general game playing. The primary advantage is that neuroevolution may be applied to a wider variety of problems than supervised learning methods, which need a curriculum of accurate input-output pairings to function properly. Neuroevolution, on the other hand, needs nothing more than a measurement of how well a network performs at a given job. For instance, the result of a game can be easily measured even if the necessary strategies are not provided in the form of labeled examples. Neuroevolution is frequently utilized as a component of the reinforcement learning paradigm. It can be contrasted with conventional deep learning approaches, which make use of gradient descent on a neural network that possesses a fixed topology. Neuroevolution is frequently utilized as a component of the reinforcement learning paradigm.

How You Will Benefit

(I) Insights, and validations about the following topics:

Chapter 1: Neuroevolution

Chapter 2: Artificial neural network

Chapter 3: Evolutionary algorithm

Chapter 4: Genetic representation

Chapter 5: Effective fitness

Chapter 6: Neuroevolution of augmenting topologies

Chapter 7: Recurrent neural network

Chapter 8: Compositional pattern-producing network

Chapter 9: HyperNEAT

Chapter 10: Evolving intelligent system

(II) Answering the public top questions about neuroevolution.

(III) Real world examples for the usage of neuroevolution in many fields.

Who This Book Is For

Professionals, undergraduate and graduate students, enthusiasts, hobbyists, and those who want to go beyond basic knowledge or information for any kind of neuroevolution.

What Is Artificial Intelligence Series

The artificial intelligence book series provides comprehensive coverage in over 200 topics. Each ebook covers a specific Artificial Intelligence topic in depth, written by experts in the field. The series aims to give readers a thorough understanding of the concepts, techniques, history and applications of artificial intelligence. Topics covered include machine learning, deep learning, neural networks, computer vision, natural language processing, robotics, ethics and more. The ebooks are written for professionals, students, and anyone interested in learning about the latest developments in this rapidly advancing field.
The artificial intelligence book series provides an in-depth yet accessible exploration, from the fundamental concepts to the state-of-the-art research. With over 200 volumes, readers gain a thorough grounding in all aspects of Artificial Intelligence. The ebooks are designed to build knowledge systematically, with later volumes building on the foundations laid by earlier ones. This comprehensive series is an indispensable resource for anyone seeking to develop expertise in artificial intelligence.

• Language: English
• Publisher: One Billion Knowledgeable
• Release date: Jun 21, 2023

Book preview

Chapter 1: Neuroevolution

Neuroevolution, often spelled neuro-evolution, is a kind of artificial intelligence that generates artificial neural networks (ANN), parameters, and rules via the application of evolutionary algorithms. Neuroevolution is also used in evolutionary robotics. The primary advantage is that neuroevolution may be applied to a wider variety of problems than supervised learning methods, which need a curriculum of accurate input-output pairings to function properly. Neuroevolution, on the other hand, needs nothing more than a measurement of how well a network performs at a given job. For instance, the result of a game, such as which player came out on top and which one came out on the losing end, may be simply quantified even in the absence of labeled instances of the desired techniques. Neuroevolution is often implemented as an element of the reinforcement learning paradigm. It may be compared with traditional deep learning approaches, which typically include gradient descent being performed on a neural network that has a predetermined topology.

There have been a lot of different neuroevolution algorithms described. The difference between algorithms that develop just the strength of the link weights for a given network topology (a kind of neuroevolution known as conventional neuroevolution in certain circles) and algorithms that evolve both the topology of the network and its weights is a frequent one (called TWEANNs, for Topology and Weight Evolving Artificial Neural Network algorithms).

Techniques that evolve the structure of ANNs in parallel to its parameters (methods that employ basic evolutionary algorithms) may be differentiated from methods that develop their parameters on their own in a separate category. This differentiation can be established (through memetic algorithms).

Instead of neuroevolution, the majority of neural networks make use of gradient descent. Researchers at Uber stated around 2017 that they had found that simple structural neuroevolution algorithms were competitive with sophisticated modern industry-standard gradient-descent deep learning algorithms. This was in part due to the fact that neuroevolution was found to be less likely to get stuck in local minima than the other algorithms. Journalist Matthew Hutson, writing for Science, presented a hypothesis in which he hypothesized that one of the reasons why neuroevolution is thriving where it had failed in the past is because of the enhanced processing capacity that is accessible in the 2010s.

In evolutionary algorithms, a population of genotypes is what they work with (also referred to as genomes). In neuroevolution, a genotype is mapped to a phenotype of a neural network, and this phenotypic is then tested on some kind of activity to determine the organism's fitness.

In encoding systems known as direct encoding, the phenotype is encoded directly from the genotype. In other words, the genotype contains direct and explicit instructions for each and every neuron and link that makes up the neural network. On the other hand, indirect encoding approaches rely on the genotype to implicitly specify how that network ought to be created.

In addition to other regularities, modularity; a reduction in the size of the search area made possible by the compression of the phenotype into a more manageable genotype; mapping the issue domain to the search space (genome), also known as.

Indirect encodings that make use of artificial embryogeny (also known as artificial development) have, historically speaking, been divided into two distinct categories: those that use a grammatical approach and those that take a cell chemistry approach. The first method creates new sets of rules by developing grammatical rewriting systems. The latter makes an effort to imitate the way that physical structures form via gene expression in biological systems. Indirect encoding systems often combine elements of the two different methods.

A taxonomy for embryogenic systems has been proposed by Stanley and Miikkulainen. This taxonomy is meant to represent the fundamental features of these systems. Any embryogenic system may be positioned along any one of the five continuous dimensions that are identified in the taxonomy:

The ultimate properties and function of a cell (neuron) in its mature phenotypic are referred to as the cell's destiny. This metric measures the number of different approaches that may be