Search

Inspired by Darwin's natural selection, genetic programming is an evolutionary computation technique which searches for computer programs best solving an optimization problem. The ability of GP to perform structural optimization at the same time of parameter optimization makes it uniquely suitable to solve more complex optimization problems, in which the structure of the solution is not known a priori. But, as GP is applied to increasingly difficult problems, the efficiency of the algorithm has been severely limited by bloating. Previous studies of bloating suggest that bloating can be resolved either directly by delaying the growth in depth and size, or indirectly by making GP to find optimal solutions faster. This thesis explores both options in order to improve the scalability and the capacity of GP algorithm. It tackles the former by firstly systematically analyzing the effect of bloating using a mathematical tool developed called activation rate. It then proposes depth difference hypothesis as a new cause of bloating and investigates depth constraint crossover as a new bloating control method, which is able to give very competitive control over bloating without affecting the exploration of fitter individuals. This thesis explores the second option by developing norm-referenced fitness function, which dynamically determines the individual's fitness based on not only how well it performs, but also the population's average performance as well. It is shown both theoretically and empirically that, norm-referenced fitness is able to significantly improve GP performance over the standard GP setup.

Directed energy deposition (DED) is an additive manufacturing process which can be used to produce customized complicated parts with less cost compared with the conventional subtractive manufacturing method. However, the parts produced by DED can be deformed severely due to excessive heat accumulation. The input energy, typically laser and electron beam, is commonly regulated in many literatures to minimize the deformation based on the temperature information of the existing clad. Although the energy input control method is proved effective in reducing the deformation, the productivity of DED process is not increased. In this thesis, a feed rate-based fuzzy logic controller is proposed to maintain geometric consistency and increase the production rate at the same time. The geometric consistency and productivity are effectively improved, through numerical simulation and experiments of thin wall and square blocks in stainless steel 316L. However, feed rate control is not functional for small parts without sufficient cooling time. Therefore, a series of square blocks with various dimensions are modeled to test the threshold for choosing the proper control method. Additionally, a python program is developed to accelerate the simulation speed.

Imagine that you are trying to read everything you got in 2011 as soon as possible. That will take the first three months of 2012. How can we get access to the vast unstructured literature, automatically process it, effectively predigest and make sense of it with less effort? Focusing on the entire preprocessing and classification steps, a hybrid semi-supervised text classification approach proposed in this dissertation will help you survive in a rising sea of information. The porter stemming, new adaptive TFIDF-LDA weighting, Zipf's law based dimension reduction, multinomial Naive Bayes classifier, and Expectation-maximization algorithm are harmoniously integrated together in the hybrid semi-supervised text classification model. From a small set of “known” labeled papers, you can use this mixture model to make predictions about newly “unknown” unclassified papers into the predefined categories. Extensive experimental results show that the proposed system dramatically reduces the feature dimension and improves the classification accuracy.

This dissertation examines topics of business cycles and consumption. The first chapter studies the investment boom in the 1990s, which was sharply reversed in 2001. Standard equilibrium business cycle models have difficulties in predicting the investment boom and overshooting. A novel embodied technology and learning model is constructed that replicates the pattern of investment boom and collapse in the 1990s. New technology is assumed to increase the productivity of capital of all vintages, but only new capital can facilitate the adoption of the new technology. In addition, agents have imperfect information about the magnitude of technology shocks. A sufficient condition for a persistent investment boom and overshooting is presented. The model is solved numerically in a dynamic general equilibrium (DGE) setup as well. The model extends the standard DGE business cycle models in two ways: first, it presents a strong internal propagation mechanism; second, it generates endogenous recessions without invoking technology regress. The second chapter presents a model addressing why the marginal propensity to consume is large for small transitory unanticipated income shocks but small for large shocks of the same type. Such relationship has been shown to be quite pronounced in empirical works but cannot be explained by the Rational Expectation Permanent Income Hypothesis (RE/PIH). The model in this chapter embeds the Cash-in-Advance and transaction cost assumptions in an otherwise standard consumption optimization model and explains the empirical findings well. At the same time the model predicts that consumers only respond to small anticipated income shocks with excess sensitivities but not to large shocks. In addition, the model sheds lights on the equity premium puzzle. The mechanism introduced in this model is a promising factor that contributes to account for the puzzle, though the mechanism alone is not able to explain the entire puzzle. The third chapter studies empirically how consumption responds to a particular type of anticipated income shocks. In the period after the last mortgage payment, homeowners experience an increase in their disposable income. The study tests whether households smooth consumption over the transition period as predicted by the RE/PIH. The study finds that households do not alter nondurable goods consumption in the period following the last mortgage payment. Instead, they increase both financial savings and savings in durable goods such as housefurnishings and entertainment equipments in the year of the last mortgage payment.