Browse
Recent Submissions
Now showing 1 - 5 of 162
- ItemTowards Effective Machine Translation For A Low-Resource Agglutinative Language: Karachay-Balkar(2022) Rice, Enora; Washington, Jonathan; Grissom, AlvinNeural machine translation (NMT) is often heralded as the most effective approach to machine translation due to its success on language pairs with large parallel corpora. However, neural methods produce less than ideal results on low-resource languages when their performance is evaluated using accuracy metrics like the Bilingual Evaluation Understudy (BLEU) score. One alternative to NMT is rule-based machine translation (RBMT), but it too has drawbacks. Furthermore, little research has been done to compare the two approaches on criteria beyond their respective accuracies. This thesis evaluates RBMT and NMT systems holistically based on efficacy, ethicality, and utility to low-resource language communities. Using the language Karachay-Balkar as a case-study, the latter half of this thesis investigates how two free and open-source machine translation packages, Apertium (rule-based) and JoeyNMT (neural), might support community-driven machine translation development. While neither platform is found to be ideal, this thesis finds that the Apertium is more conducive to a community driven machine translation development process than JoeyNMT when evaluated on the criteria of efficiency, accessibility, ease of deployment, and interpretability.
- ItemAn Introduction to Ray Tracing and its Optimization Methods with a Focus on Denoising Through Filters(2022) Huang, Haosong; Xu, DiannaRasterization and ray tracing are two common techniques used in computer graphics rendering which displays complex 3D scenes containing billions of objects on our computer. Compared to rasterization, ray tracing can generate higher-quality and even realistic image by simulating the behaviors of physical lights at the cost of tremendously higher computational time. Researchers have taken decades to increase the efficiency of ray tracing and reduce the rendering time of one image from one day to several seconds. One research goal is to investigate the possibility of real time ray tracing whose rendering time is below 1/24 second so people can see animated photo-realistic images in games or simulations. Although real time ray tracing is a hot topics in rendering, it is very difficult for beginners to understand ray tracing through reading hundreds of pages of textbooks. In this literature review we provide a summarized top-down approach to explain the basic concepts of ray tracing and some of the optimization methods, so readers may obtain some insights about ray tracing without reading too many pages of rigorous math and code. This literature review starts with the basic framework and some of the early methods of ray tracing - ranging from ray casting, recursive ray tracing, Monte Carlo ray tracing to path tracing. The second section lists some existing optimization methods to ray tracing at each stage of the framework as examples. The third section looks at a specific optimization method called denoising and gives detailed explanations to the classical filtering method of denoising using spatio-temporal variance-guided filter (SVGF) as an example. Some machine learning applications in ray tracing optimization will also be mentioned.
- ItemTopic Modeling, Named-Entity Recognition, and Network Analysis of Literary Corpora(2022) Velonis, Adrian; Boudourides, MosesIn this thesis, we conduct a literature review on the application of two natural language processing techniques, topic modeling and named-entity recognition (character identification), on collections of literary fiction. These techniques allow us to efficiently identify the dominant themes in a text as well as the placement of named entities in relation to those themes. This process can be extended to the corpus as a whole to gauge the presence of themes across multiple works. We also investigate the use of this data in networks, which allow researchers to create human-readable maps of themes and entities across the corpus.
- ItemImage Segmentations via Graph Cuts(2022) Palnitkar, Rahul; Farias Sales Rocha Neto, JeovaIn this paper, we discuss Shi and Malik's groundbreaking Normalized Cuts approach to image segmentation, and review a number of modified Normalized Cut methods, including Mixed Normalized Cut, Semi-Supervised Normalized Cuts, and the gPb contour detector. Additionally, we discuss alternate interpretations of the Normalized Cut. After a review of previous approaches to image segmentation, we propose a new method, building off of the Normalized Cuts algorithm by constructing a new image graph which holds pixel color information. We then test our new approach, Color-Nodes Segmentation, comparing its performance and accuracy to the standard Ncuts approach pioneered by Shi and Malik. Color-Nodes Segmentation offers a more accurate segmentation for noisy images, and a five-to-seven fold improvement in runtime as compared to the classical Ncuts approach.
- ItemThe Usage and Computation of Quasi-Polynomials in Optimization Techniques(2022) Rittler, Alexander; Wonnacott, David G.Many varying analyses and transformations in the field of program optimization rely on an ability to answer counting questions; specifically, how many elements satisfy a set of conditions. Thus, it is indispensable to be able to model such questions and develop a mathematics that answers them efficiently and precisely. This paper explains how quasi-polynomials are just that and what might be done to expand their use. To achieve this, we first motivate the techniques and terminology in program optimization relevant to the discussion. We then present the usefulness of quasi-polynomials in the field. Next, we define quasi-polynomials both informally and mathematically, discuss the techniques used to compute them, address their own limitations, and review a couple of their applications in detail. Finally, based on these results, we consider the problem of approximating quasi-polynomials with polynomial bounds and propose a course of inquiry for bounding the relative error of these approximations in future work.