Curiosity of human nature drives us to explore the origins of what makes each of us different. From ancient legends and mythology, Mendel's law, Punnett square to modern genetic research, we carry on this old but eternal question. Thanks to technological revolution, today's scientists try to answer this question using easily measurable gene expression and other profiling data. However, the exploration can easily get lost in the data of growing volume, dimension, noise and complexity. This dissertation is aimed at developing new machine learning methods that take data from different classes as input, augment them with knowledge of feature relationships, and train classification models that serve two goals: 1) class prediction for previously unseen samples; 2) knowledge discovery of the underlying causes of class differences. Application of our methods in genetic studies can help scientist take advantage of existing biological networks, generate diagnosis with higher accuracy, and discover the driver networks behind the differences. We proposed three new graph-based regularization algorithms. Graph Connectivity Constrained AdaBoost algorithm combines a connectivity module, a deletion function, and a model retraining procedure with the AdaBoost classifier. Graph-regularized Linear Programming Support Vector Machine integrates penalty term based on submodular graph cut function into linear classifier's objective function. Proximal Graph LogisticBoost adds lasso and graph-based penalties into logistic risk function of an ensemble classifier. Results of tests of our models on simulated biological datasets show that the proposed methods are able to produce accurate, sparse classifiers, and can help discover true genetic differences between phenotypes.

Data exchange between different information systems is a complex issue. Each system, designed for a specific purpose, is defined using a vocabulary of the specific business. While Web services allow interoperations and data communications between multiple systems, the clients of the services must understand the vocabulary of the targeting data resources to select services or to construct queries. In this thesis we explore an ontology-based approach to facilitate clients’ queries in the vocabulary of the clients’ own domain, and to automate the query processing. A governmental inter-department data query process has been used to illustrate the capability of the semantic approach.

Negation is a part of language that humans engage in pretty much from the onset of speech. Negation appears at first glance to be harder to grasp than object or action labels, yet this thesis explores how this family of ‘concepts’ could be acquired in a meaningful way by a humanoid robot based solely on the unconstrained dialogue with a human conversation partner. The earliest forms of negation appear to be linked to the affective or motivational state of the speaker. Therefore we developed a behavioural architecture which contains a motivational system. This motivational system feeds its state simultaneously to other subsystems for the purpose of symbol-grounding but also leads to the expression of the robot’s motivational state via a facial display of emotions and motivationally congruent body behaviours. In order to achieve the grounding of negative words we will examine two different mechanisms which provide an alternative to the established grounding via ostension with or without joint attention. Two large experiments were conducted to test these two mechanisms. One of these mechanisms is so called negative intent interpretation, the other one is a combination of physical and linguistic prohibition. Both mechanisms have been described in the literature on early child language development but have never been used in human-robot-interaction for the purpose of symbol grounding. As we will show, both mechanisms may operate simultaneously and we can exclude none of them as potential ontogenetic origin of negation.

This thesis presents the use of a new sigmoid activation function in backpropagation artificial neural networks (ANNs). ANNs using conventional activation functions may generalize poorly when trained on a set which includes quirky, mislabeled, unbalanced, or otherwise complicated data. This new activation function is an attempt to improve generalization and reduce overtraining on mislabeled or irrelevant data by restricting training when inputs to the hidden neurons are sufficiently small. This activation function includes a flattened, low-training region which grows or shrinks during back-propagation to ensure a desired proportion of inputs inside the low-training region. With a desired low-training proportion of 0, this activation function reduces to a standard sigmoidal curve. A network with the new activation function implemented in the hidden layer is trained on benchmark data sets and compared with the standard activation function in an attempt to improve area under the curve for the receiver operating characteristic in biological and other classification tasks.

Given the rise in the application of neural networks to all sorts of interesting problems, it seems natural to apply them to statistical tests. This senior thesis studies whether neural networks built to classify discrete circular probability distributions can outperform a class of well-known statistical tests for uniformity for discrete circular data that includes the Rayleigh Test1, the Watson Test2, and the Ajne Test3. Each neural network used is relatively small with no more than 3 layers: an input layer taking in discrete data sets on a circle, a hidden layer, and an output layer outputting probability values between 0 and 1, with 0 mapping to uniform and 1 mapping to nonuniform. In evaluating performances, I compare the accuracy, type I error, and type II error of this class of statistical tests and of the neural networks built to compete with them. 1 Jammalamadaka, S. Rao(1-UCSB-PB); SenGupta, A.(6-ISI-ASU)Topics in circular statistics. (English summary) With 1 IBM-PC floppy disk (3.5 inch; HD). Series on Multivariate Analysis, 5. World Scientific Publishing Co., Inc., River Edge, NJ, 2001. xii+322 pp. ISBN: 981-02-3778-2 2 Watson, G. S.Goodness-of-fit tests on a circle. II. Biometrika 49 1962 57–63. 3 Ajne, B.A simple test for uniformity of a circular distribution. Biometrika 55 1968 343–354.

In order for assistive robots to collaborate effectively with humans for completing everyday tasks, they must be endowed with the ability to effectively perceive scenes and more importantly, recognize human intentions. As a result, we present in this dissertation a novel scene-dependent human-robot collaborative system capable of recognizing and learning human intentions based on scene objects, the actions that can be performed on them, and human interaction history. The aim of this system is to reduce the amount of human interactions necessary for communicating tasks to a robot. Accordingly, the system is partitioned into scene understanding and intention recognition modules. For scene understanding, the system is responsible for segmenting objects from captured RGB-D data, determining their positions and orientations in space, and acquiring their category labels. This information is fed into our intention recognition component where the most likely object and action pair that the user desires is determined. Our contributions to the state of the art are manifold. We propose an intention recognition framework that is appropriate for persons with limited physical capabilities, whereby we do not observe human physical actions for inferring intentions as is commonplace, but rather we only observe the scene. At the core of this framework is our novel probabilistic graphical model formulation entitled Object-Action Intention Networks. These networks are undirected graphical models where the nodes are comprised of object, action, and object feature variables, and the links between them indicate some form of direct probabilistic interaction. This setup, in tandem with a recursive Bayesian learning paradigm, enables our system to adapt to a user's preferences. We also propose an algorithm for the rapid estimation of position and orientation values of scene objects from single-view 3D point cloud data using a multi-scale superquadric fitting approach. Additionally, we leverage recent advances in computer vision for an RGB-D object categorization procedure that balances discrimination and generalization as well as a depth segmentation procedure that acquires candidate objects from tabletops. We demonstrate the feasibility of the collaborative system presented herein by conducting evaluations on multiple scenes comprised of objects from 11 categories, along with 7 possible actions, and 36 possible intentions. We achieve approximately 81% reduction in interactions overall after learning despite changes to scene structure.

Coding Electronic Medical Records (EMRs) with diagnosis and procedure codes is an essential task for billing, secondary data analyses, and monitoring health trends. Both speed and accuracy of coding are critical. While coding errors could lead to more patient-side financial burden and misinterpretation of a patient’s well-being, timely coding is also needed to avoid backlogs and additional costs for the healthcare facility. Therefore, it is necessary to develop automated diagnosis and procedure code recommendation methods that can be used by professional medical coders. The main difficulty with developing automated EMR coding methods is the nature of the label space. The standardized vocabularies used for medical coding contain over 10 thousand codes. The label space is large, and the label distribution is extremely unbalanced - most codes occur very infrequently, with a few codes occurring several orders of magnitude more than others. A few codes never occur in training dataset at all. In this work, we present three methods to handle the large unbalanced label space. First, we study how to augment EMR training data with biomedical data (research articles indexed on PubMed) to improve the performance of standard neural networks for text classification. PubMed indexes more than 23 million citations. Many of the indexed articles contain relevant information about diagnosis and procedure codes. Therefore, we present a novel method of incorporating this unstructured data in PubMed using transfer learning. Second, we combine ideas from metric learning with recent advances in neural networks to form a novel neural architecture that better handles infrequent codes. And third, we present new methods to predict codes that have never appeared in the training dataset. Overall, our contributions constitute advances in neural multi-label text classification with potential consequences for improving EMR coding.

Electronic medical record (EMR) data is collected on a daily basis at hospitals and other healthcare facilities to track patients’ health situations including conditions, treatments (medications, procedures), diagnostics (labs) and associated healthcare operations. Besides being useful for individual patient care and hospital operations (e.g., billing, triaging), EMRs can also be exploited for secondary data analyses to glean discriminative patterns that hold across patient cohorts for different phenotypes. These patterns in turn can yield high level insights into disease progression with interventional potential. In this dissertation, using a large scale realistic EMR dataset of over one million patients visiting University of Kentucky healthcare facilities, we explore data mining and machine learning methods for association rule (AR) mining and predictive modeling with mood and anxiety disorders as use-cases. Our first work involves analysis of existing quantitative measures of rule interestingness to assess how they align with a practicing psychiatrist’s sense of novelty/surprise corresponding to ARs identified from EMRs. Our second effort involves mining causal ARs with depression and anxiety disorders as target conditions through matching methods accounting for computationally identified confounding attributes. Our final effort involves efficient implementation (via GPUs) and application of contrast pattern mining to predictive modeling for mental conditions using various representational methods and recurrent neural networks. Overall, we demonstrate the effectiveness of rule mining methods in secondary analyses of EMR data for identifying causal associations and building predictive models for diseases.

In order to reduce the time associated with and the costs of drug discovery, machine learning is being used to automate much of the work in this process. However the size and complex nature of molecular data makes the application of machine learning especially challenging. Much work must go into the process of engineering features that are then used to train machine learning models, costing considerable amounts of time and requiring the knowledge of domain experts to be most effective. The purpose of this work is to demonstrate data driven approaches to perform the feature selection and extraction steps in order to decrease the amount of expert knowledge required to model interactions between proteins and drug molecules.

The process of designing a floorplan is highly iterative and requires extensive human labor. Currently, there are a number of computer programs that aid humans in floorplan design. These programs, however, are limited in their inability to fully automate the creative process. Such automation would allow a professional to quickly generate many possible floorplan solutions, greatly expediting the process. However, automating this creative process is very difficult because of the many implicit and explicit rules a model must learn in order create viable floorplans. In this paper, we propose a method of floorplan generation using two machine learning models: a sequential model that generates rooms within the floorplan, and a graph-based model that finds adjacencies between generated rooms. Each of these models can be altered such that they are each capable of producing a floorplan independently; however, we find that the combination of these models outperforms each of its pieces, as well as a statistic-based approach.