As machine learning becomes an increasingly relevant field being incorporated into everyday life, so does the need for consistently high performing models. With these high expectations, along with potentially restrictive data sets, it is crucial to be able to use techniques for machine learning that increase the likelihood of success. Robust Principal Component Analysis (RPCA) not only extracts anomalous data, but also finds correlations among the given features in a data set, in which these correlations can themselves be used as features. By taking a novel approach to utilizing the output from RPCA, we address how our method effects the performance of such models. We take into account the efficiency of our approach, and use projectors to enable our method to have a 99.79% faster run time. We apply our method primarily to cyber security data sets, though we also investigate the effects on data sets from other fields (e.g. medical).

Microarray technology has enriched the study of gene expression in such a way that scientists are now able to measure the expression levels of thousands of genes in a single experiment. Microarray gene expression data gained great importance in recent years due to its role in disease diagnoses and prognoses which help to choose the appropriate treatment plan for patients. This technology has shifted a new era in molecular classification, interpreting gene expression data remains a difficult problem and an active research area due to their native nature of “high dimensional low sample size”. Such problems pose great challenges to existing classification methods. Thus, effective feature selection techniques are often needed in this case to aid to correctly classify different tumor types and consequently lead to a better understanding of genetic signatures as well as improve treatment strategies. This thesis aims on a comparative study of state-of-the-art feature selection methods, classification methods, and the combination of them, based on gene expression data. We compared the efficiency of three different classification methods including: support vector machines, k- nearest neighbor and random forest, and eight different feature selection methods, including: information gain, twoing rule, sum minority, max minority, gini index, sum of variances, t- statistics, and one-dimension support vector machine. Five-fold cross validation was used to evaluate the classification performance. Two publicly available gene expression data sets of glioma were used for this study. Different experiments have been applied to compare the performance of the classification methods with and without performing feature selection. Results revealed the important role of feature selection in classifying gene expression data. By performing feature selection, the classification accuracy can be significantly boosted by using a small number of genes. The relationship of features selected in different feature selection methods is investigated and the most frequent features selected in each fold among all methods for both datasets are evaluated.

Recognizing visual information is crucial for many real artificial-intelligence-based applications, ranging from domestic robots to autonomous vehicles. However, the success of deep learning methods on visual recognition tasks is highly dependent on access to large-scale labeled datasets, which are expensive and cumbersome to collect. Transfer learning provides a way to alleviate the burden of annotating data, which transfers the knowledge learned from a rich-labeled source domain to a scarce-labeled target domain. However, the performance of deep learning models degrades significantly when testing on novel domains due to the presence of domain shift. To tackle the domain shift, conventional domain adaptation methods diminish the domain shift between two domains with a distribution matching loss or adversarial loss. These models align the domain-specific feature distribution and the domain-invariant feature distribution simultaneously, which is sub-optimal towards solving deep domain adaptation tasks, given that deep neural networks are known to extract features in which multiple hidden factors are highly entangled. This thesis explores how to learn effective transferable features by disentangling the deep features. The following questions are studied: (1) how to disentangle the deep features into domain-invariant and domain-specific features? (2) how would feature disentanglement help to learn transferable features under a synthetic-to-real domain adaptation scenario? (3) how would feature disentanglement facilitate transfer learning with multiple source or target domains? (4) how to leverage feature disentanglement to boost the performance in a federated system? To address these needs, this thesis proposes deep adversarial feature disentanglement: a class/domain identifier is trained on the labeled source domain and the disentangler generates features to fool the class/domain identifier. Extensive experiments and empirical analysis demonstrate the effectiveness of the feature disentanglement method on many real-world domain adaptation tasks. Specifically, the following three unsupervised domain adaptation scenarios are explored: (1) domain agnostic learning with disentangled representations, (2) unsupervised federated domain adaptation, (3) multi-source domain adaptation.

Underwater localization is a difficult task when it comes to making the system autonomous due to the unpredictable environment. The fact that radio signals such as GPS cannot be transmitted through water makes autonomous movement and localization underwater even more challenging. One specific method that is widely used for autonomous underwater navigation applications is Simultaneous Localization and Mapping (SLAM), a technique in which a map is created and updated while localizing the vehicle within the map. In SLAM, feature detection is used in landmark extraction and data association by examining each pixel and differentiating landmarks pixels from those of the background. Previous research on the performance of different feature detection methods have been done in environments such as cisterns and caverns where the effects of the ocean are reduced. Our objective, however, is to achieves robust localization in the open ocean environment of the Cal Poly pier. This thesis performs a comparative study between different feature detection methods including Scale Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), and Oriented FAST and Rotated BRIEF (ORB) on different sensors. We evaluate the feature detection and matching performance of these algorithms in a simulated open-ocean environment.

We trained a combined deep convolutional neural network to predict seals’ age (3 categories) and gender (2 categories). The entire dataset contains 110 seals with around 489 thousand location records. Most records are continuous and measured in a certain step. We created five convolutional layers for feature representation and established two fully connected structure as age’s and gender’s classifier, respectively. Each classifier consists of three fully connected layers. Treating seals’ latitude and longitude as input, entire deep learning network, which includes 780,000 neurons and 2,097,000 parameters, can reach to 70.72% accuracy rate for predicting seals’ age and simultaneously achieve 79.95% for gender estimation.

Modern business applications are getting increasingly distributed as multi-tenant software as a service (SaaS). This leads to new challenges in terms of quality assurance, because all customers are directly affected by software changes. The resulting problem is to proactively determinate evolutionary effects. Because SaaS applications are often realized in the sense of a software product line, this thesis examines ways of using feature models to face the mentioned problem. For this purpose, two approaches are analyzed: extended feature models with quality attributes annotated per feature and the analysis of structural aspects of feature models and corresponding concrete configurations. The presented attributed feature model approach measures the quality of concrete configurations to make configurations comparable according to specific quality goals. Criteria are elicited for when configurations can be compared to draw helpful conclusions. The structural approach focuses economic questions that are quality assurance related, such as identifying features that none of the tenants selected in their application configurations. Furthermore, three algorithms are presented that demonstrate the structural analysis approach to gather information relevant to quality assurance.

We present a framework for extraction and prediction of online workload data from a workload manager of a mainframe operating system. To boost overall system performance, the prediction will be corporated into the workload manager to take preventive action before a bottleneck develops. Model and feature selection automatically create a prediction model based on given training data, thereby keeping the system flexible. We tailor data extraction, preprocessing and training to this specific task, keeping in mind the nonstationarity of business processes. Using error measures suited to our task, we show that our approach is promising. To conclude, we discuss our first results and give an outlook on future work.

Quality inspection is an indispensable part of the production process of screws for hardware manufactories. In general, hardware manufactories do the quality test of screws by using an electric screwdriver to twist screws. However, there are some limitations and shortcomings in the manual inspection. Firstly, the efficiency of manual inspection is low. Second, manual inspection is difficult to achieve continuous working for 24 hours, which will make a high wage cost. In this thesis, in order to enhance the inspection efficiency and save test costs, we propose to use the image recognition technology of memristor neural networks to check the quality of screws. Here, we discuss different training models of neural networks, namely: convolutional neural networks, one-layer memristor neural network with fixed learning rates. By using the dataset of 8,202 screw head images, experimental results show that the classification accuracy of CNNs and memristor neural networks can achieve 96% and 90%, respectively, which prove the effectiveness of the proposed method.

Understanding eyewitness lineup judgment processes is critical, both from a theoretical standpoint (to better understand human memory) and from a practical one (to prevent wrongful convictions and criminals walking free). Currently, two influential theories attempt to explain lineup decision making: the theory of eyewitness judgment strategies (Lindsay & Wells, 1985), and the signal detection theory-informed diagnostic-feature-detection hypothesis (Wixted & Mickes, 2014). The theory of eyewitness judgment strategies posits that eyewitnesses can adopt either an absolute judgment strategy (base identification decisions only on their memory for the culprit) or a relative judgment strategy (base identification decision on lineup member comparisons). This theory further predicts that relative judgment strategies lead to an increase in false identifications. Contrast this with the diagnostic-feature-detection hypothesis, which predicts that the lineup member comparisons inherent to relative strategies promote greater accuracy. These two theories have been tested indirectly (i.e., via lineup format manipulations tangentially related to the theory), but there is a lack of direct tests. Across two experiments (Ns = 192, 584), we presented participants with simulated crime videos and corresponding lineups, and manipulated judgment strategy using explicit absolute and relative strategy instructions and a novel rank-order manipulation meant to encourage lineup member comparisons. We found no substantial differences in identifications or overall accuracy as a function of instructed strategy. These results are inconsistent with the theory of eyewitness judgment strategies but provide some support for the diagnostic-feature-detection hypothesis. We discuss implications for both theories and future lineup research. / Graduate

Against biblical odds, the baseball version of a stage mom and her grifting ex road trip their kids’ baseball team across the midwest in the name of life, liberty, and the little league world series.