Background: Cannabis use (CU) and disorders (CUD) are highly prevalent among adolescents and young adults. We aim to identify clinically meaningful latent classes of users of cannabis and other illegal substances with distinct problem profiles. Methods: N= 3021 community subjects aged 14–24 at baseline were followed-up over a period ranging up to 10 years. Substance use (SU) and disorders (SUD) were assessed with the DSM-IV/M-CIDI. Latent class analysis (LCA) was conducted with a subset of N= 1089 subjects with repeated illegal SU. The variables entered in the LCA were CU-related problems, CUD, other SUD, and other mental disorders. Results: Four latent classes were identified: “Unproblematic CU” (class 1: 59.2%), “Primary alcohol use disorders” (class 2: 14.4%), “Delinquent cannabis/alcohol DSM-IV-abuse” (class 3: 17.9%), “CUD with multiple problems” (class 4: 8.5%). Range and level of CU-related problems were highest in classes 3 and 4. Comorbidity with other mental disorders was highest in classes 2 and 4. The probability of alcohol disorders and unmet treatment needs was considerable in classes 2–4. Conclusion: While the majority of subjects with repeated illegal SU did not experience notable problems over the 10-year period, a large minority (40.8%) experienced problematic outcomes, distinguished by clinically meaningful profiles. The data underline the need for specifically tailored interventions for adolescents with problematic CU and highlight the potentially important role of alcohol and other mental disorders.

This study was designed to empirically derive latent classes of mothers who were sexually abused during childhood and to assess the association between depression, alcohol/drug use, supportive intimate partner, and specific classes. One hundred six women between the ages of 20 and 44 years (M = 27) who reported having been sexually abused during childhood (CSA) and 158 non-CSA mothers between the ages of 20 and 43 years (M = 23) were interviewed and assessed along six parenting dimensions. Logistic regression models evaluated the association between psychoemotional variables and specific classes. The final model consisted of three classes--53.2%, 31.7%, and 15.2%. Alcohol/drug use was not statistically significantly associated with either class. Maternal depression and intimate partner support were differentially associated with the three parenting classes. Empirical support is provided for distinct classes of mothers sexually abused in childhood. The data-driven categorization of CSA mothers provides research and clinical directions for future parenting of survivors of childhood sexual abuse.

Reinforcement Learning has applications in various domains, but the typical assumption is of a stationary process. Hence, when this hypothesis does not hold, performance may be sub-optimal. Tackling non-stationarity is not a trivial task because it requires adaptation to changing environments and predictability in various conditions, as dynamics and rewards might change over time. Meta Reinforcement Learning has been used to handle the non-stationary evolution of the environment while knowing the potential source of noise in the system. However, our research presents a novel method to manage such complexity by learning a suitable latent representation that captures relevant patterns for decision-making, improving the policy optimization procedure. We present a two-step framework that combines the unsupervised training of Deep Variational Auto-encoders to extract latent variables and a state-of-the-art model-free and off-policy Batch Reinforcement Learning algorithm called Fitted Q-Iteration, without relying on any assumptions about the environment dynamics. This framework is named Latent-Variable Fitted Q-Iteration (LV-FQI). Furthermore, to validate the generalization and robustness capabilities for exploiting the structure of the temporal sequence of time-series data and extracting near-optimal policies, we evaluated the performance with empirical experiments on synthetic data generated from classical financial models. We also tested it on Foreign Exchange trading scenarios with various degrees of non-stationarity and low signal-to-noise ratios. The results showed performance improvements compared to existing algorithms, indicating great promise for addressing the long-standing challenges of Continual Reinforcement Learning. / Reinforcement Learning har tillämpningar inom olika områden, men den typiska antagningen är att det rör sig om en stationär process. När detta antagande inte stämmer kan prestationen bli suboptimal. Att hantera icke-stationaritet är ingen enkel uppgift eftersom det kräver anpassning till föränderliga miljöer och förutsägbarhet under olika förhållanden, då dynamiken och belöningarna kan förändras över tiden. Meta Reinforcement Learning har använts för att hantera den icke-stationära utvecklingen av miljön genom att känna till potentiella källor till brus i systemet. Vår forskning presenterar emellertid en ny metod för att hantera en sådan komplexitet genom att lära en lämplig latent representation som fångar relevanta mönster för beslutsfattande och förbättrar optimeringsprocessen för policyn. Vi presenterar en tvåstegsramverk som kombinerar osuperviserad träning av Deep Variational Auto-encoders för att extrahera latenta variabler och en state-of-the-art model-free och off-policy Batch Reinforcement Learning-algoritm, Fitted Q-Iteration, utan att förlita sig på några antaganden om miljöns dynamik. Detta ramverk kallas Latent-Variable Fitted Q-Iteration (LV-FQI). För att validera generaliserings- och robusthetsförmågan att utnyttja strukturen hos den tidsmässiga sekvensen av tidsseriedata och extrahera nära-optimala policys utvärderade vi prestandan med empiriska experiment på syntetiska data genererade från klassiska finansiella modeller. Vi testade också det på handelsscenario för Foreign Exchange med olika grader av icke-stationaritet och låg signal-till-brus-förhållande. Resultaten visade prestandaförbättringar jämfört med befintliga algoritmer och indikerar stor potential för att tackla de långvariga utmaningarna inom kontinuerlig Reinforcement Learning.

<p dir="ltr">Ice thermal energy storage (ITES) systems have long been an economic way to slash cooling costs in the commercial sector since the 1980s. An ITES system generates cooling in the formation of ice within a storage tank. This occurs during periods of the day when the cost of electricity is low, normally at night. This ice is then melted to absorb the energy within the conditioned space. While ITES systems have been prosperous in the commercial sector, they have yet to take root in the residential sector.</p><p dir="ltr">The U.S. Department of Energy (DoE) has published guidelines for TES. The DoE guidelines include providing a minimum of four hours of cooling, shifting 30-50% of a space’s cooling load to non-peak hours, minimizing the weight, volume, complexity, and cost of the system, creating a system than operates for over 10,000 cycles, enacting predictive control measures, and being modular to increase scale for larger single-family and multi-family homes [1]. The purpose of this research is to develop a model that meets these guidelines.</p><p dir="ltr">After extensive research in both experimental data, technical specifications, existing models, and best practices taken from the works of others a MATLAB model was generated. The modeled ITES system is comprised of a 1m diameter tank by 1m tall. Ice was selected as the PCM. A baseline model was constructed with parameters deemed to be ideal. This model generated an ITES system that can be charged in under four hours and is capable of providing a total of 22.18 kWh of cooling for a single-family home over a four-hour time period. This model was then validated with experimental data and found to have a root mean squared error of 0.0959 for the system state of charge. During the validation both the experimental and model estimation for the water/ice within the tank converged at the HTF supply temperature of -5.2°C.</p><p dir="ltr">With the model established, a parametric analysis was conducted to learn how adjusting a few of the system parameters impact it. The first parameter, reducing the pipe radius, has the potential to lead to a 152.6-minute reduction in charge time. The second parameter, varying the heat transfer fluid (HTF) within the prescribed zone of 0.7 kg/s to 1.2 kg/s, experienced a 4.8-minute increase in charge time for the former and a decrease in charge time by 5.4 minutes for the latter. The third parameter, increasing the pipe spacing and consequently increasing the ratio of mass of water to mass of HTF, yielded a negative impact. A 7.1mm increase in pipe spacing produced a 16.6-minute increase in charge time. Meanwhile, a 14.2mm increase in pipe spacing created a 93.3-minute increase in charge time and exceeded the charging time limit of five hours.</p><p dir="ltr">This functioning model establishes the foundation of creating a residential-scale ITES system. The adjustability and scalability of the code enable it to be modified to user specifications. Thus, allowing for various prototypes to be generated based on it. The model also lays the groundwork to synthesize a code containing an ITES system and a heat pump operating as one. This will aid in the understanding of residential-scale ITES systems and their energy effects.</p>

Master of Science / Department of Computing and Information Sciences / Doina Caragea / Several computational biology and bioinformatics problems involve DNA sequence classification using supervised machine learning algorithms. The performance of these algorithms is largely dependent on the availability of labeled data and the approach used to represent DNA sequences as {\it feature vectors}. For many organisms, the labeled DNA data is scarce, while the unlabeled data is easily available. However, for a small number of well-studied model organisms, large amounts of labeled data are available. This calls for {\it domain adaptation} approaches, which can transfer knowledge from a {\it source} domain, for which labeled data is available, to a {\it target} domain, for which large amounts of unlabeled data are available. Intuitively, one approach to domain adaptation can be obtained by extracting and representing the features that the source domain and the target domain sequences share. \emph{Latent Dirichlet Allocation} (LDA) is an unsupervised dimensionality reduction technique that has been successfully used to generate features for sequence data such as text. In this work, we explore the use of LDA for generating predictive DNA sequence features, that can be used in both supervised and domain adaptation frameworks. More precisely, we propose two dimensionality reduction approaches, LDA Words (LDAW) and LDA Distribution (LDAD) for DNA sequences. LDA is a probabilistic model, which is generative in nature, and is used to model collections of discrete data such as document collections. For our problem, a sequence is considered to be a ``document" and k-mers obtained from a sequence are ``document words". We use LDA to model our sequence collection. Given the LDA model, each document can be represented as a distribution over topics (where a topic can be seen as a distribution over k-mers). In the LDAW method, we use the top k-mers in each topic as our features (i.e., k-mers with the highest probability); while in the LDAD method, we use the topic distribution to represent a document as a feature vector. We study LDA-based dimensionality reduction approaches for both supervised DNA sequence classification, as well as domain adaptation approaches. We apply the proposed approaches on the splice site predication problem, which is an important DNA sequence classification problem in the context of genome annotation. In the supervised learning framework, we study the effectiveness of LDAW and LDAD methods by comparing them with a traditional dimensionality reduction technique based on the information gain criterion. In the domain adaptation framework, we study the effect of increasing the evolutionary distances between the source and target organisms, and the effect of using different weights when combining labeled data from the source domain and with labeled data from the target domain. Experimental results show that LDA-based features can be successfully used to perform dimensionality reduction and domain adaptation for DNA sequence classification problems.

Master of Science / Department of Computing and Information Sciences / Doina Caragea / Gurdip Singh / In the recent years, social networks have become an integral part of our lives. Their outgrowth has resulted in opportunities for interesting data mining problems, such as interest or friendship recommendations. A global ontology over the interests specified by the users of a social network is essential for accurate recommendations. The focus of this work is on engineering such an interest ontology. In particular, given that the resulting ontology is meant to be used for data mining applications to social network problems, we explore only hierarchical ontologies. We propose, evaluate and compare three approaches to engineer an interest hierarchy. The proposed approaches make use of two popular knowledge bases, Wikipedia and Directory Mozilla, to extract interest definitions and/or relationships between interests. More precisely, the first approach uses Wikipedia to find interest definitions, the latent semantic analysis technique to measure the similarity between interests based on their definitions, and an agglomerative clustering algorithm to group similar interests into higher level concepts. The second approach uses the Wikipedia Category Graph to extract relationships between interests. Similarly, the third approach uses Directory Mozilla to extract relationships between interests. Our results indicate that the third approach, although the simplest, is the most effective for building an ontology over user interests. We use the ontology produced by the third approach to construct interest based features. These features are further used to learn classifiers for the friendship prediction task. The results show the usefulness of the ontology with respect to the results obtained in absence of the ontology.

Reactivation of latent infection is the major cause of tuberculosis (TB). Cholesterol is a critical carbon source during latent infection. Catabolism of cholesterol contributes to the pool of propionyl-CoA, a precursor that is incorporated into cell-wall lipids. Arylamine N-acetyltransferase (NAT) is encoded within a gene cluster that is involved in the sterol-ring degradation and is essential for intracellular survival. NAT from M. tuberculosis (TBNAT) can utilise propionyl-CoA and therefore was proposed as a target for TB-drug development. Deleting the nat gene or inhibiting the NAT enzyme prevents intracellular survival and results in depletion of cell-wall lipids. NAT inhibitors, including the piperidinol class, were identified by high-throughput screening. The insolubility of recombinant TBNAT has been a major limitation in pursuing it as a drug target. Subcloning tbnat into a pVLT31 vector resulted in a yield of 6-16 mg/litre-bacterial-culture of pure-soluble recombinant TBNAT. The increased yield allowed for extensive screening for crystallisation conditions. However, since a structure was not obtained, the model NAT from M. marinum (MMNAT) was employed to further understand NAT as a target. Screening against a panel of Acyl-CoA cofactors showed that MMNAT can also utilise propionyl-CoA. The MMNAT structure in complex with the high affinity substrate hydralazine was determined (2.1 Å) and the architecture of the arylamine pocket was delineated. A novel mechanism for the acetylation reaction of hydralazine has emerged. It is proposed that the acetyl group is transferred from acetyl-CoA to the heterocyclic aromatic nitrogen of hydralazine, which explains the immediate cyclisation of the acetylated metabolite into an N-methyltriazolophthalazine. By employing mass spectroscopy, enzyme assays, computational docking and structural studies, a covalent mechanism of inhibition by the piperidinol class was established, and the inhibitor-binding pocket was identified. Inhibitors with new scaffolds were identified using the in silico 3D-shape screening and thermal shift assay.

In this thesis we provide statistical and model-based data mining methods for pattern detection with applications to biomedical and healthcare data sets. In particular, we examine applications in costly acute or chronic disease management. In Chapter II, we consider nuclear magnetic resonance experiments in which we seek to locate and demix smooth, yet highly localized components in a noisy two-dimensional signal. By using wavelet-based methods we are able to separate components from the noisy background, as well as from other neighboring components. In Chapter III, we pilot methods for identifying profiles of patient utilization of the healthcare system from large, highly-sensitive, patient-level data. We combine model-based data mining methods with clustering analysis in order to extract longitudinal utilization profiles. We transform these profiles into simple visual displays that can inform policy decisions and quantify the potential cost savings of interventions that improve adherence to recommended care guidelines. In Chapter IV, we propose new methods integrating survival analysis models and clustering analysis to profile patient-level utilization behaviors while controlling for variations in the population’s demographic and healthcare characteristics and explaining variations in utilization due to different state-based Medicaid programs, as well as access and urbanicity measures.

The purpose of this Clinic project is to help Expedia, Inc. expand the search capabilities it offers to its users. In particular, the goal is to help the company respond to unconstrained search queries by generating a method to associate hotels and regions around the world with the higher-level attributes that describe them, such as “family- friendly” or “culturally-rich.” Our team utilized machine-learning algorithms to extract metadata from textual data about hotels and cities. We focused on two machine-learning models: decision trees and Latent Dirichlet Allocation (LDA). The first appeared to be a promising approach, but would require more resources to replicate on the scale Expedia needs. On the other hand, we were able to generate useful results using LDA. We created a website to visualize these results.

Diplodia pinea causes Diplodia tip blight on more than 30 different pine species. During the past 10 years, Diplodia tip blight has emerged as a serious problem in landscape and Christmas tree farms in this region. Surveys of diseased and symptomless Austrian pines revealed that latent infections of symptomless shoots by D. pinea were common. Latent infections may account for the recently observed rapid decline of mildly diseased pines in our region. To investigate the colonization habits of D. pinea within its host, molecular cytology was attempted and traditional histology was performed on naturally infected, diseased and asymptomatic Austrian pine tissues. I devoted much effort to developing a transformation system for D. pinea. Ultimately I did not succeed in this goal, but I was able to develop a highly efficient protocol for Agrobacterium tumefaciens-mediated transformation of another pathogenic fungus, Colletotrichum graminicola, in the process. The work that I did should help in future efforts to transform D. pinea, something that will be essential if it is to become a tractable system for the study of fungal latency. Traditional histological methods were more successful, and provided important information about the nature of latent infections. Very sparse epiphytic and subcuticular fungal growth was observed in healthy shoots, however, no fungal tissues were present within the shoots. In diseased and latently infected shoots, crevices created between the needle bundles and the shoots were filled with fungal material, and hyphae were observed colonizing the needle sheaths. Hyphae were also observed breaching the shoot epidermal layer in these crevices and colonizing the underlying periderm. D. pinea colonization was extensive in all tissues of diseased shoots early in symptom development. In contrast, localized pockets of degradation were observed in the periderm and adjacent cortical cells located around areas of needle attachment in asymptomatic, latently infected shoots. The mechanism that operates to prevent expansion of these infected pockets in the latently infected shoots is still unclear. Obvious signs of pine defense mechanisms were only observed in 2 shoots. My observations were consistent with the idea that colonization progresses into the vascular tissues, and that this results in symptom development. Vascular colonization may occur more readily if the host is stressed. My research lays the groundwork for future efforts to understand the nature of the transformation from latent to pathogenic infection.