[pt] Métodos de previsão são muito importantes para o desenvolvimento de diversas atividades no cotidiano produtivo de nossa sociedade. Vários modelos estatísticos são desenvolvidos até hoje, contudo possuem muitos pressupostos que devem ser seguidos com o intuito de se obter uma resposta aceitável. Modelos não estatísticos para prever séries temporais como os que envolvem Sistemas de Inferência Fuzzy (SIFs) fornecem uma descrição do processo de previsão por meio de regras linguísticas. Explora-se, nesta dissertação, o GPFIS-Forecast: versão do GPFIS - Sistema de Inferência Fuzzy Genético baseado em Programação Genética Multigênica - para previsão de série temporais univariadas. O modelo apresenta, em sua execução, quatro etapas básicas: Fuzzificação, Inferência, Defuzzificação e Avaliação. Em cada uma destas etapas, pode-se fazer uso de diferentes configurações, com implicações evidentes nos resultados finais. Este trabalho propõe o aperfeiçoamento do GPFIS-Forecast em duas vertentes principais:( i) aumentar a quantidade de possibilidades de configurações, avaliando se podem contribuir significativamente para a acurácia das previsões;(ii) adicionar informações complementares como alternativas para a interpretação do resultado do modelo, tendo como compromisso tanto a acurácia e quanto a interpretabilidade. Os estudos de caso demonstram que, em casos de séries temporais com pouca tendência, o GPFIS-Forecast apresenta uma acurácia entre as 10 melhores da competição NN3; quando há forte tendência, faz-se necessário o uso de pré-processamento, prejudicando a interpretabilidade do resultado. Os Limites de Previsão Fuzzy introduzidos neste trabalho agregam mais informação ao resultado da previsão pontual, apontando possíveis ajustes finais nas bases de regras de modelos com maior granularidade. / [en] Forecasting methods are very important for the development of various activities in everyday society. Several statistical models have been developed, but many assumptions must be made in order to obtain an acceptable response. Nonstatistical models for time series forecasting such as those involving systems Fuzzy Inference Systems (FIS) provide a description of the process through linguistic rules. This dissertation delves into GPFISForecast: a version of GPFIS - Fuzzy Inference System based on Multigene Genetic Programming - for univariate time series forecasting. This model consists of four basic stages: Fuzzification, Inference, Defuzzification and Evaluation. In each of these steps, different configurations will have distinct impacts on the results. This work proposes the improvement of GPFIS-Forecast along two main lines (i) increase the amount of possible configurations and assess their contribution to a better forecasting accuracy and (ii) add further information to the interpretation of results, keeping in mind both accuracy and interpretability. The case studies show that in the case of time series with small tendency, GPFIS-Forecast provides a good accuracy; when tendency is larger and pre-processing becomes necessary, interpretability is affected. The Fuzzy Forecasting Limits introduced here add more information to the result, pointing to possible adjustments to rule bases of models with greater granularity.

The high social costs associated with bankruptcy have spurred searches for better theoretical understanding and prediction capability. In this paper, we investigate a hybrid approach to bankruptcy prediction, using a genetic programming algorithm to construct a bankruptcy prediction model with variables from a rough sets model derived in prior research. Both studies used data from 291 US public companies for the period 1991 to 1997. The second stage genetic programming model developed in this research consists of a decision model that is 80% accurate on a validation sample as compared to the original rough sets model which was 67% accurate. Additionally, the genetic programming model reveals relationships between variables that are not apparent in either the rough sets model or prior research. These findings indicate that genetic programming coupled with rough sets theory can be an efficient and effective hybrid modeling approach both for developing a robust bankruptcy prediction model and for offering additional theoretical insights.

This Master's thesis presents a novel approach for detecting and predicting attack vectors based on genetic programming. The proposed method utilizes a genetic algorithm to evolve a set of rules that predict attack vectors over the system based on caught indicators of compromise. The generated rules are then used to identify potential attack vectors and predict how it started and how it will develop in future. The research aims to improve the accuracy and efficiency of existing methods for attack detection and prediction. The proposed approach is evaluated using real-world attack data and compared against several state-of-the-art techniques. Results indicate that the proposed method outperforms existing approaches in terms of detection accuracy and prediction capability. This research has important implications for the field of cybersecurity and can assist organizations in developing more effective and proactive defense strategies against cyberattacks. Background. Cybersecurity is an increasingly critical issue in today's digital age. Cyberattacks are becoming more sophisticated, making it challenging for traditional defense mechanisms to detect and prevent them. Therefore, it is crucial to develop new and innovative methods for identifying and predicting potential attack vectors. In this context, this Master's thesis presents a novel approach to detecting and predicting attack vectors based on genetic programming. The proposed method aims to improve the accuracy and efficiency of existing approaches to cyberattack detection and prediction. Objectives.This Master’s thesis aims to reach the following objectives: 1. To identify the limitations of existing approaches to cyberattack detection and prevention and propose a novel method based on genetic programming. 2. To develop a genetic programming-based algorithm to evolve a model for attack-vectors prediction. 3. To evaluate the effectiveness of the proposed approach using real-world attack data Methods. The methods used in this Master's thesis combine literature review, data collection, algorithm development, experimentation, data analysis, and recommendations to improving approach to detecting and predicting attack vectors using genetic programming. The research aims to contribute to the field of cybersecurity by advancing our understanding of cyberattack detection and prevention. Results. The proposed method has the potential to enhance the accuracy and efficiency of cyberattack detection and prediction, which can help organizations prevent or mitigate the impact of cyberattacks. Future improvements can include more complex MITRE ATT&CK datasets, including Mobile and ICS matrices. Conclusions. The genetic programming-based algorithm developed in this thesis was shown to be effective in detecting and predicting attack vectors using real-world attack data. The proposed approach has the potential to improve organizations' cybersecurity posture by providing a proactive defense strategy against cyberattacks.

[pt] Esta dissertação propõe modelos parcimoniosos para tarefas de regressão e classificação em conjuntos de dados exclusivamente seccionais, mantendo-se a hipótese de amostragem aleatória. Os modelos de regressão são lineares, estimados por Mínimos Quadrados Ordinários resolvidos pela Decomposição QR, apresentando solução única sob posto cheio ou não da matriz de regressores. Os modelos de classificação são não lineares, estimados por Máxima Verossimilhança utilizando uma variante do Método de Newton, nem sempre apresentando solução única. A parcimônia dos modelos de regressão é fundamentada na prova matemática de que somente agregará acurácia ao modelo o regressor que apresentar módulo da estatística de teste, em um teste de hipótese bicaudal, superior à unidade. A parcimônia dos modelos de classificação é fundamentada em significância estatística e embasada intuitivamente no resultado teórico da existência de classificadores perfeitos. A Programação Genética (PG) realiza o processo de evolução de modelos, explorando o espaço de busca de possíveis modelos, constituídos de distintos regressores. Os resultados obtidos via Programação Genética Econométrica (PGE) – nome dado ao algoritmo gerador de modelos – foram comparados aos proporcionados por benchmarks em oito distintos conjuntos de dados, mostrando-se competitivos em termos de acurácia na maior parte dos casos. Tanto sob o domínio da PG quanto sob o domínio da econometria, a PGE mostrou benefícios, como o auxílio na identificação de introns, o combate ao bloat por significância estatística e a geração de modelos econométricos de elevada acurácia, entre outros. / [en] This dissertation proposes parsimonious models for regression and classification tasks in cross-sectional datasets under random sample hypothesis. Regression models are linear in parameters, estimated by Ordinary Least Squares solved by QR Decomposition, presenting a unique solution under full rank of the regressor matrix or not. Classification models are nonlinear in parameters, estimated by Maximum Likelihood, not always presenting a unique solution. Parsimony in regression models is based on the mathematical proof that accuracy will be added to models only by the regressor that presents a test statistic module higher than a predefined value in a two-sided hypothesis test. Parsimony in classification models is based on statistical significance and, intuitively, on the theoretical result about the existence of perfect classifiers. Genetic Programming performs the evolution process of models, being responsible for exploring the search space of possible regressors and models. The results obtained with Econometric Genetic Programming – name of the algorithm in this dissertation – was compared with those from benchmarks in eight distinct cross-sectional datasets, showing competitive results in terms of accuracy in most cases. Both in the field of Genetic Programming and in that of econometrics, Econometric Genetic Programming has shown benefits such as help on introns identification, combat to bloat by statistical significance and generation of high level accuracy models, among others.

We investigate an approach to apply Genetic Programming for the evolution of optimal escaping strategies of a team of caribou agents in the wolf-caribou predator prey problem (WCPPP) where the WCPPP is comprised of a team of caribou agents attempting to escape from a single yet superior (in terms of sensory abilities, raw speed, and maximum energy) wolf agent in a simulated twodimensional infinite toroidal world. We empirically verify our hypothesis that the incorporation of empathy in caribou agents significantly improves both the evolution efficiency of the escaping behavior and the effectiveness of such a behavior. This finding may be viewed as a verification of the survival value of empathy and the resulting compassionate behavior of the escaping caribou agents. Moreover, considering the fact that a single caribou cannot escape from the superior wolf, the ability of a team of empathic caribou agents to escape may also be viewed as an illustration of the emergent nature of a successful escaping behavior – in that the team-level properties are more than the mere sum of the properties of the individual entities. Within this context, we also present empirical results that verify the complex (nonlinear) nature of the relationship between the size of team of caribou agents and the efficiency of their escaping behavior. / 博士(工学) / Doctor of Philosophy in Engineering / 同志社大学 / Doshisha University

[pt] A Programação Genética permite que computadores resolvam problemas automaticamente, sem que eles tenham sido programados para tal. Utilizando a inspiração no princípio da seleção natural de Darwin, uma população de programas, ou indivíduos, é mantida, modificada baseada em variação genética, e avaliada de acordo com uma função de aptidão (fitness). A programação genética tem sido usada com sucesso por uma série de aplicações como projeto automático, reconhecimento de padrões, controle robótico, mineração de dados e análise de imagens. Porém, a avaliação da gigantesca quantidade de indivíduos gerados requer excessiva quantidade de computação, levando a um tempo de execução inviável para problemas grandes. Este trabalho explora o alto poder computacional de unidades de processamento gráfico, ou GPUs, para acelerar a programação genética e permitir a geração automática de programas para grandes problemas. Propomos duas novas metodologias para se explorar a GPU em programação genética: compilação em linguagem intermediária e a criação de indivíduos em código de máquina. Estas metodologias apresentam vantagens em relação às metodologias tradicionais usadas na literatura. A utilização de linguagem intermediária reduz etapas de compilação e trabalha com instruções que estão bem documentadas. A criação de indivíduos em código de máquina não possui nenhuma etapa de compilação, mas requer engenharia reversa das instruções que não estão documentadas neste nível. Nossas metodologias são baseadas em programação genética linear e inspiradas em computação quântica. O uso de computação quântica permite uma convergência rápida, capacidade de busca global e inclusão da história passada dos indivíduos. As metodologias propostas foram comparadas com as metodologias existentes e apresentaram ganhos consideráveis de desempenho. Foi observado um desempenho máximo de até 2,74 trilhões de GPops (operações de programação genética por segundo) para o benchmark Multiplexador de 20 bits e foi possível estender a programação genética para problemas que apresentam bases de dados de até 7 milhões de amostras. / [en] Genetic Programming enables computers to solve problems automatically, without being programmed to it. Using the inspiration in the Darwin s Principle of natural selection, a population of programs or individuals is maintained, modified based on genetic variation, and evaluated according to a fitness function. Genetic programming has been successfully applied to many different applications such as automatic design, pattern recognition, robotic control, data mining and image analysis. However, the evaluation of the huge amount of individuals requires excessive computational demands, leading to extremely long computational times for large size problems. This work exploits the high computational power of graphics processing units, or GPUs, to accelerate genetic programming and to enable the automatic generation of programs for large problems. We propose two new methodologies to exploit the power of the GPU in genetic programming: intermediate language compilation and individuals creation in machine language. These methodologies have advantages over traditional methods used in the literature. The use of an intermediate language reduces the compilation steps, and works with instructions that are well-documented. The individuals creation in machine language has no compilation step, but requires reverse engineering of the instructions that are not documented at this level. Our methodologies are based on linear genetic programming and are inspired by quantum computing. The use of quantum computing allows rapid convergence, global search capability and inclusion of individuals past history. The proposed methodologies were compared against existing methodologies and they showed considerable performance gains. It was observed a maximum performance of 2,74 trillion GPops (genetic programming operations per second) for the 20-bit Multiplexer benchmark, and it was possible to extend genetic programming for problems that have databases with up to 7 million samples.

To create a realistic environment, many simulations require simulated agents with human behavior patterns. Manually creating such agents with realistic behavior is often a tedious and time-consuming task. This dissertation describes a new approach that automatically builds human behavior models for simulated agents by observing human performance. The research described in this dissertation synergistically combines Context-Based Reasoning, a paradigm especially developed to model tactical human performance within simulated agents, with Genetic Programming, a machine learning algorithm to construct the behavior knowledge in accordance to the paradigm. This synergistic combination of well-documented AI methodologies has resulted in a new algorithm that effectively and automatically builds simulated agents with human behavior. This algorithm was tested extensively with five different simulated agents created by observing the performance of five humans driving an automobile simulator. The agents show not only the ability/capability to automatically learn and generalize the behavior of the human observed, but they also capture some of the personal behavior patterns observed among the five humans. Furthermore, the agents exhibited a performance that was at least as good as agents developed manually by a knowledgeable engineer.

Earth ecosystems and environment have been changing rapidly due to the advanced technologies and developments of humans. Impacts caused by human activities and developments are difficult to acquire for evaluations due to the rapid changes. Remote sensing (RS) technology has been implemented for environmental managements. A new and promising trend in remote sensing for environment is widely used to measure and monitor the earth environment and its changes. RS allows large-scaled measurements over a large region within a very short period of time. Continuous and repeatable measurements are the very indispensable features of RS. Soil moisture is a critical element in the hydrological cycle especially in a semiarid or arid region. Point measurement to comprehend the soil moisture distribution contiguously in a vast watershed is difficult because the soil moisture patterns might greatly vary temporally and spatially. Space-borne radar imaging satellites have been popular because they have the capability to exhibit all weather observations. Yet the estimation methods of soil moisture based on the active or passive satellite imageries remain uncertain. This study aims at presenting a systematic soil moisture estimation method for the Choke Canyon Reservoir Watershed (CCRW), a semiarid watershed with an area of over 14,200 km2 in south Texas. With the aid of five corner reflectors, the RADARSAT-1 Synthetic Aperture Radar (SAR) imageries of the study area acquired in April and September 2004 were processed by both radiometric and geometric calibrations at first. New soil moisture estimation models derived by genetic programming (GP) technique were then developed and applied to support the soil moisture distribution analysis. The GP-based nonlinear function derived in the evolutionary process uniquely links a series of crucial topographic and geographic features. Included in this process are slope, aspect, vegetation cover, and soil permeability to compliment the well-calibrated SAR data. Research indicates that the novel application of GP proved useful for generating a highly nonlinear structure in regression regime, which exhibits very strong correlations statistically between the model estimates and the ground truth measurements (volumetric water content) on the basis of the unseen data sets. In an effort to produce the soil moisture distributions over seasons, it eventually leads to characterizing local- to regional-scale soil moisture variability and performing the possible estimation of water storages of the terrestrial hydrosphere. A new evolutionary computational, supervised classification scheme (Riparian Classification Algorithm, RICAL) was developed and used to identify the change of riparian zones in a semi-arid watershed temporally and spatially. The case study uniquely demonstrates an effort to incorporating both vegetation index and soil moisture estimates based on Landsat 5 TM and RADARSAT-1 imageries while trying to improve the riparian classification in the Choke Canyon Reservoir Watershed (CCRW), South Texas. The CCRW was selected as the study area contributing to the reservoir, which is mostly agricultural and range land in a semi-arid coastal environment. This makes the change detection of riparian buffers significant due to their interception capability of non-point source impacts within the riparian buffer zones and the maintenance of ecosystem integrity region wide. The estimation of soil moisture based on RADARSAT-1 Synthetic Aperture Radar (SAR) satellite imagery as previously developed was used. Eight commonly used vegetation indices were calculated from the reflectance obtained from Landsat 5 TM satellite images. The vegetation indices were individually used to classify vegetation cover in association with genetic programming algorithm. The soil moisture and vegetation indices were integrated into Landsat TM images based on a pre-pixel channel approach for riparian classification. Two different classification algorithms were used including genetic programming, and a combination of ISODATA and maximum likelihood supervised classification. The white box feature of genetic programming revealed the comparative advantage of all input parameters. The GP algorithm yielded more than 90% accuracy, based on unseen ground data, using vegetation index and Landsat reflectance band 1, 2, 3, and 4. The detection of changes in the buffer zone was proved to be technically feasible with high accuracy. Overall, the development of the RICAL algorithm may lead to the formulation of more effective management strategies for the handling of non-point source pollution control, bird habitat monitoring, and grazing and live stock management in the future. Soil properties, landscapes, channels, fault lines, erosion/deposition patches, and bedload transport history show geologic and geomorphologic features in a variety of watersheds. In response to these unique watershed characteristics, the hydrology of large-scale watersheds is often very complex. Precipitation, infiltration and percolation, stream flow, plant transpiration, soil moisture changes, and groundwater recharge are intimately related with each other to form water balance dynamics on the surface of these watersheds. Within this chapter, depicted is an optimal site selection technology using a grey integer programming (GIP) model to assimilate remote sensing-based geo-environmental patterns in an uncertain environment with respect to some technical and resources constraints. It enables us to retrieve the hydrological trends and pinpoint the most critical locations for the deployment of monitoring stations in a vast watershed. Geo-environmental information amassed in this study includes soil permeability, surface temperature, soil moisture, precipitation, leaf area index (LAI) and normalized difference vegetation index (NDVI). With the aid of a remote sensing-based GIP analysis, only five locations out of more than 800 candidate sites were selected by the spatial analysis, and then confirmed by a field investigation. The methodology developed in this remote sensing-based GIP analysis will significantly advance the state-of-the-art technology in optimum arrangement/distribution of water sensor platforms for maximum sensing coverage and information-extraction capacity. Effective water resources management is a critically important priority across the globe. While water scarcity limits the uses of water in many ways, floods also have caused so many damages and lives. To more efficiently use the limited amount of water or to resourcefully provide adequate time for flood warning, the results have led us to seek advanced techniques for improving streamflow forecasting. The objective of this section of research is to incorporate sea surface temperature (SST), Next Generation Radar (NEXRAD) and meteorological characteristics with historical stream data to forecast the actual streamflow using genetic programming. This study case concerns the forecasting of stream discharge of a complex-terrain, semi-arid watershed. This study elicits microclimatological factors and the resultant stream flow rate in river system given the influence of dynamic basin features such as soil moisture, soil temperature, ambient relative humidity, air temperature, sea surface temperature, and precipitation. Evaluations of the forecasting results are expressed in terms of the percentage error (PE), the root-mean-square error (RMSE), and the square of the Pearson product moment correlation coefficient (r-squared value). The developed models can predict streamflow with very good accuracy with an r-square of 0.84 and PE of 1% for a 30-day prediction.

Return Oriented Programming (ROP) is the de-facto technique used to exploit most of today’s native-code vulnerabilities hiding in old and newly developed software alike. By reusing bits and pieces of already existing code (gadgets), ROP can be used to bypass the ever-present Write ⊕ eXecute (W⊕X) security feature, which enforces memory to only be marked as either executable or writable; never both at the same time. Even with its widespread use, crafting more advanced ROP-chains is mostly left as a manual task. This paper attempts to explore the viability of automating ROP-chain generation by leveraging genetic programming (GP), and describes the implementation and design of the ROP-compiler GENROP in this endeavour. We introduce a novel approach to adapt GP to work within the environment of ROP, which attempts to guide the algorithm and preemptively remove pathways which are known ahead of time to be unable to generate a solution. GENROP is tested by attempting to generate a working payload against a number of binaries, and is then evaluated based on success rate and payload size when compared to angrop (another ROP-compiler). The results show that the algorithm is able to generate functioning payloads in most of the tested cases, although it does perform worse than angrop. This can partly be explained by the fact that GENROP uses gadget definitions generated by angrop, which reduces the potential viability of the ROP-compiler, as more unwieldy but potentially usable gadgets are not available. Additionally, it was found that extensively guiding the algorithm has negative consequences in terms of solution diversity. Relying on faster execution times and more iterations might produce better results. Further work is required to assess whether or not generating ROP-chains using genetic programming is a viable approach.

Reinforced concrete block shear walls (RCBSWs) are used as seismic force resisting systems in low- and medium-rise buildings. However, attributed to their nonlinear behavior and composite material nature, accurate prediction of their seismic performance relying only on mechanics is challenging. This study introduces multi-gene genetic programming (MGGP)— a class of bio-inspired artificial intelligence, to uncover the complexity of RCBSW behaviors and develop simplified procedures for predicting the full backbone curve of flexure-dominated fully grouted RCBSWs under cyclic loading. A piecewise linear backbone curve was developed using five secant stiffness expressions associated with cracking, yielding, 80% ultimate, ultimate, and 20% strength degradation (i.e., post-peak stage) derived through controlled MGGP. Based on the experimental results of large-scale cyclically loaded RCBSWs, compiled from previously reported studies, a variable selection procedure was performed to identify the most influential variable subset governing wall behaviors. Utilizing individual wall results, the MGGP stiffness expressions were first trained and tested, and their accuracy was subsequently compared to that of existing models employing various statistical measures. In addition, the predictability of the developed backbone model was assessed at the system-level against experimental results of two two-story buildings available in the literature. The outcomes obtained from this study demonstrate the power of MGGP approach in addressing the complexity of the cyclic behavior of RCBSWs at both component- and system-level—offering an efficient prediction tool that can be adopted by relevant seismic design standards pertaining to RCBSW buildings. / Thesis / Master of Applied Science (MASc)