Genetic Programming: Fundamentals and Applications

By Fouad Sabry

What Is Genetic Programming

In the field of artificial intelligence, genetic programming, abbreviated GP, is a method for making a population of unsuited programs evolve into programs that are more suited to a certain task. This is accomplished by applying operations that are equivalent to natural genetic processes to the population of programs.

How You Will Benefit

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

Chapter 1: Genetic programming

Chapter 2: Bio-inspired computing

Chapter 3: CMA-ES

Chapter 4: Fitness approximation

Chapter 5: Gene expression programming

Chapter 6: Genetic representation

Chapter 7: Grammatical evolution

Chapter 8: Inductive programming

Chapter 9: Linear genetic programming

Chapter 10: Multi expression programming

(II) Answering the public top questions about genetic programming.

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

(IV) 17 appendices to explain, briefly, 266 emerging technologies in each industry to have 360-degree full understanding of genetic programming' technologies.

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 genetic programming.

• Language: English
• Publisher: One Billion Knowledgeable
• Release date: Jul 1, 2023

Book preview

Chapter 1: Genetic programming

Genetic programming, often known as GP, is a method used in artificial intelligence for the purpose of developing programs in order to make them more suitable for a certain job. This is accomplished by applying operations akin to natural genetic processes to a population of programs.

The procedures consist of selecting the programs that are most suitable for reproduction (crossover) and mutation based on a predetermined fitness measure, which is often an individual's level of expertise in the desired activity. In the process of producing new and distinct children that will become a part of the next generation of programs, the crossover operation includes exchanging random pieces of chosen pairings, which are referred to as the parents. The process of changing a program by mutation entails changing one portion of the program at random with another component of the program at random. Some of the programs that were considered but ultimately rejected for replication have been transferred from the previous generation to the new generation. After that, the selection along with the other operations are applied in a recursive manner to the next generation of programs.

The members of each new generation are often healthier on average than the members of the generation that came before them, and the best-of-generation program from one generation is typically superior than the best-of-generation programs from prior generations. The evolution will often come to an end after one of its constituent programs achieves a level of competence or fitness that was previously set.

It is possible, and it really does happen rather often, that a specific execution of the algorithm will result in an early convergence to some local maximum that is neither a globally optimum nor even a decent solution. The production of a really excellent outcome often requires a number of iterations ranging from dozens to hundreds. In order to prevent diseases, it is possible that it is important to have a high beginning population size as well as heterogeneity among the individuals.

It is likely that Alan Turing's suggestion, which was made in 1950, was the first record of the idea to develop programs. showed 77 results in which Genetic Programming competed well against humans.

In 1996, Koza established the Genetic Programming conference as an annual event.

The early work that laid the groundwork for contemporary research subjects and applications in genetic programming is rather broad. Some examples of this work include software synthesis and repair, predictive modeling, and data mining, The evolution of computer programs, which are often stored in memory as tree structures, is accomplished via GP. Evaluating data in a recursive fashion is simple to do using trees. Because every node in the tree may perform an operator function, and any node, even terminal nodes, can act as an operand, developing and evaluating mathematical expressions is a simple process. Therefore, historically, GP prefers to make use of programming languages that naturally include tree structures (for example, Lisp; other functional programming languages are also suitable).

Non-tree representations have been proposed and have been successfully implemented, such as linear genetic programming, which is suited to the more conventional imperative languages [see, for example, Banzhaf et al. (1998)]. Another kind of genetic programming known as GP is called cartesian genetic programming, and it encrypts computer programs using a graph representation rather than the standard tree-based representation.

The majority of representations contain code that is not structurally effective (introns). It could seem as if such non-coding genes are pointless since they don't have any impact on the way any one person performs. They do, however, change the probability of creating distinct offspring when subjected to the variation operators, which in turn changes the variational features of the individual. Experiments seem to demonstrate quicker convergence when utilizing program representations that enable such non-coding genes, in comparison to program representations that do not contain any non-coding genes at all.

The term selection refers to the process of choosing some members of the current generation to become the parents of the subsequent generation. The selection of the people is done in a probabilistic manner in such a way that the persons who have performed better have a greater probability of being picked. Several other treatments have been shown to be more effective for a variety of general practitioner issues.

Elitism, which includes seeding the next generation with the best person (or best n people) from the present generation, is a strategy that is often used to prevent regression. Elitism entails seeding the next generation with the best individual (or best n individuals).

In the process of genetic programming, two healthy people from the population are selected to become the parents of one or two offspring. Within the framework of tree genetic programming, their parents are shown as inverted lisp-like trees, with the root nodes of their trees located at the very top. When doing a subtree crossover, a subtree from each parent is selected at random. (This will be highlighted in yellow when you watch the animation.) To create a new child tree, the subtree that was selected is removed from the parent that is donating its root (as seen on the left in the animation), and it is then replaced with a copy of the subtree that was picked at random from the other parent.

There are situations when a crossover for two children is employed, In this scenario, the removed subtree (shown on the left in the animation) is not simply discarded; rather, it is transferred to a copy of the second parent (shown on the right), where it takes the place of the randomly selected subtree in the copy.

Therefore, this kind of subtree crossover utilizes two healthy trees in order to produce two offspring trees.

Genetic programming subtree crossover

In genetic programming, there is a wide variety of mutagenesis available. They begin with a syntactically correct parent that is fit and then attempt to generate a syntactically correct offspring using randomization. During the animation, a branching subtree is selected at random (highlighted by yellow). It is eliminated, and in its place, a randomly created subtree is inserted.

Other mutation operators choose a leaf (an external node) of the tree and replace it with a leaf picked at random. This happens when the leaf is selected. Another kind of mutation involves picking a function or internal node at random and then substituting it with a different function that has the same arity (number of inputs). The Hoist mutation will choose a subtree at random and then replace it with another subtree located inside itself. As a result, the offspring of a hoist mutation will always be more petite. Replacement of the leaf and same-arity functions guarantees that the offspring will have the same dimensions