Multivariate Gaussians: We provide sample complexity upper bounds for semi-agnostically learning multivariate Gaussians under the constraint of approximate differential privacy. These are the first  finite sample upper bounds for general Gaussians which do not impose restrictions on the parameters of the distribution. Our bounds are near-optimal in the case when the covariance is known to be the identity, and conjectured to be near-optimal in the general case. From a technical standpoint, we provide analytic tools for arguing the existence of global "locally small" covers from local covers of the space. These are exploited using modifications of recent techniques for for differentially private hypothesis selection. Mixtures of Gaussians: We consider the problem of learning mixtures of Gaussians under the constraint of approximate differential privacy. We provide the first sample complexity upper bounds for privately learning mixtures of unbounded axis-aligned (or even unbounded univariate) Gaussians. To prove our results, we design a new technique for privately learning mixture distributions. A class of distributions F is said to be list-decodable if there is an algorithm that, given "heavily corrupted" samples from a distribution f in F, outputs a list of distributions, H, such that one of the distributions in H approximates f. We show that if F is privately list-decodable then we can privately learn mixtures of distributions in F. Finally, we show axis-aligned Gaussian distributions are privately list-decodable, thereby proving mixtures of such distributions are privately learnable. / Thesis / Master of Science (MSc) / Is it possible to estimate an unknown probability distribution given random samples from it? This is a fundamental problem known as distribution learning (or density estimation) that has been studied by statisticians for decades, and in recent years has become a topic of interest for computer scientists. While distribution learning is a mature and well understood problem, in many cases the samples (or data) we observe may consist of sensitive information belonging to individuals and well-known solutions may inadvertently result in the leakage of private information. In this thesis we study distribution learning under the assumption that the data is generated from high-dimensional Gaussians (or their mixtures) with the aim of understanding how many samples an algorithm needs before it can guarantee a good estimate. Furthermore, to protect against leakage of private information, we consider approaches that satisfy differential privacy — the gold standard for modern private data analysis.

Research has shown that sexual content is prevalent in television programming and that this content can have an effect on the viewer's attitudes about sex. This study examined this relationship within the theoretical framework of cultivation and social cognitive theories. This study used a survey to examine these relationships in young students at a large southeastern university. The researchers targeted freshmen, many in their first semester, for this study to get the best measure of attitudes, before they were influenced by college life. In addition to examining the effects of television viewing, the researcher looked at the effects of exposure to other media that may contain sexual content, such as magazines, Internet, DVD's, video games, and music. The study did not find significant relationships between television viewing and students' sexual attitudes and behaviors, but it did find several strong relationships among sexual attitudes and behaviors and students' exposure to other forms of media that may contain sexual content, including music, films, video games, and DVDs.

Multimedia learning tools have the potential to benefit instructors and learners as supplemental learning materials. However, when such tools are designed inappropriately, this can increase cognitive taxation and impede learning, rendering the tools ineffective. Guided by the theoretical underpinnings provided by cognitive load theory and the cognitive theory of multimedia learning, this study sought to empirically evaluate the effectiveness of a multimedia simulation tool aimed at teaching immunology to novices in an instructional setting. The instructional mode and pace of the tool were manipulated, the three levels of each variable yielding nine experimental groups. The effects of mode and pace on workload and learning scores were observed. The results of this study did not support the theory-driven hypotheses. No significant learning gains were found between the configuration groups, however overall significant learning gains were subsequently found when disregarding mode and pace configuration. Pace was found to influence workload such that fast pace presentations significantly increased workload ratings and a significant interaction of mode and pace was found for workload ratings. The findings suggest that the learning material was too high in intrinsic load and the working memory of the learners too highly taxed for the benefits of applying the design principles to be observed. Results also illustrate a potential exception to the conditions of the design principles when complex terminology is to be presented. Workload findings interpreted in the context of stress adaptation potentially indicate points at which learners at maximum capacity begin to exhibit performance decrements.

There are many different training interventions that can be used in simulation based training systems (e.g., cueing, hinting, highlighting, deliberate practice, etc.). However, the most widely used training intervention in the military is feedback, most often presented in the form of a debrief. With advances in technology, it is possible to measure and diagnose performance in real-time. Thus it is possible to provide immediate feedback during scenarios. However, training systems designers should not consider the timing of feedback in isolation. There are other parameters of feedback that must also be considered which may have an impact on performance. Specifically, feedback content and modality may also have an impact on the appropriate timing of feedback and its’ effectiveness in simulation training environments. Moreno and Mayer (2000) propose a cognitive theory of multimedia learning which describes how instruction is perceived and processed by a trainee. Using this theoretical framework, I investigate the optimal use of feedback while considering the interaction of feedback timing, content, and modality in scenario-based training environments. In order to investigate the relationship between the timing, modality, and content of feedback, a 2 (immediate, delayed) X 2 (visual, auditory) X 2 (process, outcome) betweensubjects design was used (a no feedback control condition was also included). Ninety participants were randomly assigned to the nine experimental groups. These participants performed a visual-spatial military task called the Forward Observer PC-based Simulation. Results indicated that receiving feedback was beneficial to improve performance as compared to receiving no feedback. As hypothesized, during a visual-spatial task, auditory feedback presented during a scenario led to higher performance than visual feedback. Finally, iv while I did not support my hypothesis that an interaction between all three components of feedback would affect performance, it is promising that the pattern of results mirrored the hypothesized pattern. Theoretical and practical implications, as well as limitations of the current study and directions for future research are discussed.

No description available.

The purpose of this study was to explore the ways in which high school violin students transform and comprehend a teacher’s model through the framework of Kolb’s (1984) Experiential Learning Theory. Additionally, the Learning Styles Inventory (LSI) results of this study’s participants were compared to extant researchers’ LSI data. Kolb and Kolb’s Learning Styles Inventory 3.1 and a brief survey were administered to participants (N = 100) during Phase I of the study in order to gather data regarding three quantitative independent variables—learning style, gender, and grade level. A subset of participants (n = 15) representing an array of those variables participated in Phase II of the study wherein participants were shown a video recorded lesson. During that lesson, a teacher modeled an eight-measure melodic phrase for each participant whose responses to the lesson were recorded for later analysis. Qualitative student responses (applied strategy, focus during the lesson, type of response to the model, performance intensity, task complexity) and interview responses were coded and distilled into common themes and compared among independent variables from Phase I. The high school violinist participants in this study preferred reflective observation and concrete experience orientations more frequently than was the case in extant research. The only significant interaction between independent variables was found between gender and learning preference. The two most frequently applied strategies were derived from Initiating (AE/CE = 24.46%) and Creating (CE/RO = 20.44%) learning styles. Participants largely focused on musical components (77.55%)—e.g., rhythm, pitch, intonation, articulation—by performing with the bow (48.72%) concurrently with the model (57.95%). When provided with practice time, participants largely utilized low (33. 68%) or silent, reflective (24.47%) intensities. Fundamental, two-phase combinations of strategies were applied the majority of the time (57.72%) by participants. Qualitative descriptions of the variety of participant responses were included and contextualized using LSI data. I concluded, based on a synthesis of the quantitative data and qualitative observations, that participants largely prioritized immediate individual needs—such as pitch identification or previous sections of the lesson—over both teacher instruction and their own learning preferences. I also concluded that a single modeling experience often resulted in a diverse array of participant responses—which may or may not adhere to the immediate content of the lesson. As a result of this study, I suggest that music educators and researchers consider that learners potentially utilize a singular modeling experience in a variety of different ways resulting in an array of potential outcomes. It is important for teachers to be explicit and clear in their instructions surrounding a modeling task in order to better guide students towards desired outcomes. Future researchers might consider learners’ viewpoints in response to a modeled experience as a means of framing achievement, outcome, or other research topics. ELT researchers might consider building on the implications of the comparison among KLSI data and qualitative data among learners under the age of 19 with a focus on variables outside the typical factors of gender, age, educational level, educational specialization, and culture.

The instrumental misuse of prescription stimulants as "study drugs", particularly by college students, is a serious issue that needs to be further investigated. Using data from a sample of 549 University of Central Florida Orlando students, the current study tested the relationship between prescription stimulant misuse and social learning theory, as well as general strain theory. Approximately 17% of participants reported misusing prescription stimulants for academic purposes at least once in the past year. Findings show support for social learning theory; the number of friends who use prescription stimulants and the individual's attitudes about the effectiveness of the drugs are both significant variables. General strain theory was divided into two parts; the first one tests the relationship between strain and negative affect, while the second tests negative affect, strain and prescription stimulant misuse. Overall, general strain theory was not found to explain prescription stimulant misuse. Also, results showed students who binge drink or use other substances are significantly more likely to report prescription stimulant misuse. Potential implications for these findings are discussed, as well as future research directions.

Let F and H be two (compatible) classes of functions. We observe that even when both F and H have small capacities as measured by their uniform covering numbers, the capacity of the composition class H o F={h o f| f in F, h in H} can become prohibitively large or even unbounded. To this end, in this thesis we provide a framework for controlling the capacity of composition and extend our results to bound the capacity of neural networks. Composition of Random Classes: We show that adding a small amount of Gaussian noise to the output of cF before composing it with H can effectively control the capacity of H o F, offering a general recipe for modular design. To prove our results, we define new notions of uniform covering number of random functions with respect to the total variation and Wasserstein distances. The bounds for composition then come naturally through the use of data processing inequality. Capacity of Neural Networks: We instantiate our results for the case of sigmoid neural networks. We start by finding a bound for the single-layer noisy neural network by estimating input distributions with mixtures of Gaussians and covering them. Next, we use our composition theorems to propose a novel bound for the covering number of a multi-layer network. This bound does not require Lipschitz assumption and works for networks with potentially large weights. Empirical Investigation of Generalization Bounds: We include preliminary empirical results on MNIST dataset to compare several covering number bounds based on their suggested generalization bounds. To compare these bounds, we propose a new metric (NVAC) that measures the minimum number of samples required to make the bound non-vacuous. The empirical results indicate that the amount of noise required to improve over existing uniform bounds can be numerically negligible. The source codes are available at https://github.com/fathollahpour/composition_noise / Thesis / Master of Science (MSc) / Given two classes of functions with bounded capacity, is there a systematic way to bound the capacity of their composition? We show that this is not generally true. Capacity of a class of functions is a learning-theoretic quantity that may be used to explain its sample complexity and generalization behaviour. In other words, bounding the capacity of a class can be used to ensure that given enough samples, with high probability, the deviation between training and expected errors is small. In this thesis, we show that adding a small amount of Gaussian noise to the output of functions can effectively control the capacity of composition, introducing a general framework for modular design. We instantiate our results for sigmoid neural networks and derive capacity bounds that work for networks with large weights. Our empirical results show that the amount of Gaussian noise required to improve over existing bounds is negligible.

No description available.

No description available.