Thesis (M.S.)--Bowling Green State University, 2009. / Document formatted into pages; contains xi, 81 p. : ill., maps. Includes bibliographical references.

This paper is about the possibility to use evolution to make a StarCraft AI better in some areas by using genetic programming. We aimed to use genetic programming to evolve the numbers of squad units, bunkers and turrets, which are an important part of a successful StarCraft AI. We have built a separate application for handling the evolution. This application runs in parallel with StarCraft and modifies files based on the data recieved from a played game. This is good for safety, since if StarCraft crashes the evolution is just stalled not lost. Our tests ran over the course of a few weeks. A combination of a relatively small amount of time, for something very time-consuming, and a lack of experience with genetic programming resulted in a small amount of results. The conclusion is that it is possible to improve an StarCraft AI with genetic programming, however it takes a lot of time. / Denna uppsats handlar om möjligheten att använda evolution att göra en StarCraft AI bättre i vissa områden med hjälp av genetisk programmering. Vi siktade på att använda genetisk programmering att utveckla antalet trupp enheter, bunkrar och torn, som är en viktig del av en framgångsrik StarCraft AI.

Behavior Trees are a method for AI programming that consists of a tree of hierarchical nodes controlling the flow of agent's decision making. They have proven, while being a pretty straightforward means to implement an AI, to be incredibly powerful way of obtaining autonomous agents, both due to a fact that the development can be iterable (one can start with implementing simple behavior and gradually improve the tree by adding and modifying nodes and branches) and allowing for, so to say, ``fallback tactics'', should the currently executed action fail. Born in the game industry, they have since gained fair amount of popularity in other domains, including robotics.Evolutionary algorithms, largely popularized by John Holland, have been adapted for use in a vast variety of different problems, including optimization issues and decision handling, often through introducing serious changes to both the algorithm structure and data structures used. Arguably, one of the most valuable modifications was Genetic Programming, popularized through works of John Koza. This thesis documents the work on combining Behavior Trees and Genetic Programming in order to study and observe cooperative and adversative behaviors between agents controlled by genetically generated Behavior Trees. Evolving two kinds of agents in two contrasting scenarios, this thesis focuses on feasibility of selfishness versus utilitarian behaviors and their evolution. After defining what constitutes a success for each case, we attempt to compare the results from respective scenarios to see which behavior type is profitable to exhibit.

Feature engineering is a process that augments the feature vector of a machine learning model with calculated values that are designed to enhance the accuracy of a model’s predictions. Research has shown that the accuracy of models such as deep neural networks, support vector machines, and tree/forest-based algorithms sometimes benefit from feature engineering. Expressions that combine one or more of the original features usually create these engineered features. The choice of the exact structure of an engineered feature is dependent on the type of machine learning model in use. Previous research demonstrated that various model families benefit from different types of engineered feature. Random forests, gradient-boosting machines, or other tree-based models might not see the same accuracy gain that an engineered feature allowed neural networks, generalized linear models, or other dot-product based models to achieve on the same data set. This dissertation presents a genetic programming-based algorithm that automatically engineers features that increase the accuracy of deep neural networks for some data sets. For a genetic programming algorithm to be effective, it must prioritize the search space and efficiently evaluate what it finds. This dissertation algorithm faced a potential search space composed of all possible mathematical combinations of the original feature vector. Five experiments were designed to guide the search process to efficiently evolve good engineered features. The result of this dissertation is an automated feature engineering (AFE) algorithm that is computationally efficient, even though a neural network is used to evaluate each candidate feature. This approach gave the algorithm a greater opportunity to specifically target deep neural networks in its search for engineered features that improve accuracy. Finally, a sixth experiment empirically demonstrated the degree to which this algorithm improved the accuracy of neural networks on data sets augmented by the algorithm’s engineered features.

Theme of this thesis is practical realization of so-called Island model which is one of way of parallel genetic programming. First part is theoretical. This part is describing terms of genetic programming, age-layered population structure and island model. In second part of thesis is described realization of island model in Java language.

Bankruptcy is a highly significant worldwide problem with high social costs. Traditional bankruptcy risk models have been criticized for falling short with respect to bankruptcy theory building due to either modeling assumptions or model complexity. Genetic programming minimizes the amount of a priori structure that is associated with traditional functional forms and statistical selection procedures, but still produces easily understandable and implementable models. Genetic programming was used to analyze 28 potential bankruptcy variables found to be significant in multiple prior research studies, including 10 fraud risk factors. Data was taken from a sample of 422 bankrupt and non-bankrupt Norwegian companies for the period 1993-1998. Six variables were determined to be significant. A genetic programming model was developed for the six variables from an expanded sample of 1136 bankrupt and non-bankrupt Norwegian companies. The model was 81% accurate on a validation sample, slightly better than prior genetic programming research on US public companies, and statistically significantly better than the 77% accuracy of a traditional logit model developed using the same variables and data. The most significant variable in the final model was the prior auditor opinion, thus validating the information value of the auditor's report. The model provides insight into the complex interaction of bankruptcy related factors, especially the effect of company size. The results suggest that accounting information, including the auditor's evaluation of it, is more important for larger than smaller firms. It also suggests that for small firms the most important information is liquidity and non-accounting information. The genetic programming model relationships developed in this study also support prior bankruptcy research, including the finding that company size decreases bankruptcy risk when profits are positive. It also confirms that very high profit levels are associated with increased bankruptcy risk even for large companies an association that may be reflecting the potential for management to be "Cooking the Books".

The evolution of explicitly represented topologies such as graphs involves devising methods for mutating, comparing and combining structures in meaningful ways and identifying and maintaining the necessary topological diversity. Research has been conducted in the area of the evolution of trees in genetic programming and of neural networks and some of these problems have been addressed independently by the different research communities. In the domain of neural networks, NEAT (Neuroevolution of Augmenting Topologies) has shown to be a successful method for evolving increasingly complex networks. This system's success is based on three interrelated elements: speciation, marking of historical information in topologies, and initializing search in a small structures search space. This provides the dynamics necessary for the exploration of diverse solution spaces at once and a way to discriminate between different structures. Although different representations have emerged in the area of genetic programming, the study of the tree representation has remained of interest in great part because of its mapping to programming languages and also because of the observed phenomenon of unnecessary code growth or bloat which hinders performance. The structural similarity between trees and neural networks poses an interesting question: Is it possible to apply the techniques from NEAT to the evolution of trees and if so, how does it affect performance and the dynamics of code growth? In this work we address these questions and present analogous techniques to those in NEAT for genetic programming.

Unbonded post-tensioned concrete block (UPCB) shear walls are an effective seismic force resisting system due to their ability to contain expected damage attributed to their self-centering capabilities. A few design procedures were proposed to predict the in-plane flexural response of UPCB walls, albeit following only basic mechanics and/or extensive iterative methods. Such procedures, however, may not be capable of capturing the complex nonlinear relationships between different parameters that affect UPCB walls’ behavior or are tedious to be adopted for design practice. In addition, the limited datasets used to validate these procedures may render their accuracy and generalizability questionable, further hindering their adoption by practitioners and design standards. To address these issues, an experimentally-validated nonlinear numerical model was adopted in this study and subsequently employed to simulate 95 UPCB walls with different design parameters to compensate for the lack of relevant experimental data in the current literature. Guided by mechanics and using this database, an evolutionary algorithm, multigene genetic programming (MGGP), was adopted to uncover the relationships controlling the response of UPCB walls, and subsequently develop simplified closed-form wall behavior prediction expressions. Specifically, through integrating MGGP and basic mechanics, a penta-linear backbone model was developed to predict the load-displacement backbone for UPCB walls up to 20% strength degradation. Compared to existing predictive procedures, the prediction accuracy of the developed model and its closed-form nature are expected to enable UPCB wall adoption by seismic design standards and code committees. / Thesis / Master of Applied Science (MASc)

GP has traditionally been implemented in LISP but there is a slow migration towards faster languages like C++. Any implementation language is dictated not only by the speed of the platform but also by the desirability of such an implementation. With a large number of scientists migrating to scientifically-biased programming languages like Mathematica, such provides an ideal testbed for GP.In this study it was attempted to implement GP on a Mathematica platform, exploiting the advantages of Mathematica's unique capabilities. Wherever possible, optimizations have been applied to drive the GP algorithm towards realistic goals. At an early stage it was noted that the standard GP algorithm could be significantly speeded up by parallelisation and the distribution of processing. This was incorporated into the algorithm, using known techniques and Mathematica-specific knowledge.

This diploma thesis deal with tree genetic programming algorithm. This idea is applied for solving symbolic regression tasks as well designs image filters. At first are introduced a basic concept of genetic programming and reduction of solution space. The next part presents own implementation and achieved results. Result of this work is modular system for making image filters define by specific parameters.