Browsing by Subject "Neural Networks"

Now showing 1 - 5 of 5
  • No Thumbnail Available
    Item
    Accessing The Development Of The Horizontal Translation Visual Literacy Skill In Students Using Neural Networks
    (2022-09) Andrade, Vanessa A; Prat-Resina, Xavier; Terrell, Cassidy R; Morin, Chloé S; Contreras Vital, Juquila I
    Research shows that many students struggle with visual representations in molecular life science courses, which impacts their ability to learn the material. Currently, there is minimal research related to the development of students’ visual literacy neural networks. This study aims to understand how students’ neural networks evolve based on their chemistry and biochemistry course enrollment. For this study, students as well as experts took a survey in which their horizontal translation visual literacy skills were measured to make neural networks. Then, the students’ neural networks were analyzed across the chemistry and biochemistry curricula and compared to experts’ neural networks in order to answer the question: How do the neural networks of students change throughout the different curricula in comparison to experts? Utilizing Pathfinder, eccentricity values for each node were generated in which a low value signifies a node(s) is the most central node(s). Furthermore, the degree values indicate which node(s) has the highest degree of branching. With both these values, they can be used to look at whether the neural network of students are similar or different from the experts. These data could help create a curriculum for chemistry and biochemistry courses that could possibly improve students’ visual literacy skills.
  • Thumbnail Image
    Item
    A geospatial analysis of West Nile virus in the Twin Cities metropolitan area of Minnesota.
    (2009-07) Ghosh, Debarchana
    The West Nile virus (WNV) is an infectious disease transmitted to humans and other mammals by mosquitoes that acquire the virus by feeding on WNV-infected birds. Since its initial occurrence in New York in 1999, the virus has spread rapidly west and south, causing seasonal epidemics and illness among thousands of birds, animals, and humans. Yet, we only have a rudimentary understanding of how the mosquito-borne virus operates in complex avian-human-environmental systems. The virus first reached Minnesota in 2002 and resulted in several hotspots by 2003. The year 2007 saw one of the severest incidences of WNV in Minnesota. For my dissertation research, I have developed novel approaches to understand the spread and dynamics of the virus by using key environmental, built environment, and anthropogenic risk factors that determine why, when, and where WNV strikes in the Twin Cities Metropolitan area (TCMA). The first study demonstrates the use of a novel spatiotemporal approach to identify exposure areas. The method retrospectively delineates transmission cycles as exposure areas in their entirety, involving dead birds, mosquito pools, and human cases. Given the strong spatial clustering of WNV infections in the urban areas of TCMA, the next study explores how urban landscape features contributed to the viral activities. This investigation contributed to the broader research question in the field of health geography, of how the heterogeneous urban landscape affects human health and disease patterns. The remaining studies focus on the building and interpreting a nonlinear model which captures the complex relationships between the disease incidences and the hypothesized risk factors. The goal of these studies is to identify risk factor(s) whose management would result in effective disease prevention and containment. This dissertation has applied contributions to the vector control policies. The findings from the studies can answer two fundamental questions to eliminate larva and adult mosquitoes capable of carrying WNV. First, when is the optimal time to apply insecticides and pesticides? Second, where (area) should we target spraying of pesticides? This will lead to efficient allocation of resources and allow a balance between mosquito eradication and environmental conservation efforts with respect to insecticide usage.
  • Thumbnail Image
    Item
    Optimization of Constrained Random Verification using Machine Learning
    (2018-05) Ambalakkat, Sarath Mohan
    Constrained random simulations play a critical role in Design Verification today. But the effort and time spent to manually update the input constraints, analyzing and prioritizing the unverified features in the design, significantly affect the time taken to converge to the coverage goal. This research work focuses on the optimization of constrained random verification using Machine Learning algorithms, in a coverage-driven simulation using a Universal Verification Methodology (UVM) framework. The optimization will greatly reduce the time a simulation takes to converge to the coverage goal. This research work targets automating the update of the constraints during runtime, abstracting the need for understanding the design to verify it, using Machine Learning. The verification environment is further optimized using techniques including Objective Function, Rewinding and Dynamic Seed Manipulation. The enhanced environment resolves the limitations of the previous efforts at employing these techniques, optimizing the scalability of the environment and enhancing its compatibility at verifying complex combinational designs and sequential designs including Finite State Machines (FSMs). The optimized verification environment comprises of a SystemVerilog testbench which interfaces and interacts with a TCL environment. The methodology has been empirically demonstrated, with remarkable results showing its superior quality in terms of faster automated coverage closure, efficient final stimulus solution and proposed higher quality of coverage. Multiple Machine Learning algorithms, including a Linear Regression Model and Artificial Neural Networks, have been employed to scale the compatibility of the verification environment, making it capable of autonomously verifying designs of varied behavior. Adequate simulation results to demonstrate the same have been presented in the report.
  • Thumbnail Image
    Item
    Towards Learning Powerful Deep Graph Neural Networks and Embeddings
    (2020-06) Verma, Saurabh
    Learning powerful data embeddings has recently become the core of machine learning algorithms especially in natural language processing and computer vision domains. In the graph domain, the applications of learning graph embeddings are vast and have distinguished use-cases across multi-cross domains such as bioinformatics, chemoinformatics, social networks and recommendation systems. To date, graph remains the most fundamental data structure that can represent many forms of real-world datasets. However, due to its rich but complex data structure, graph presents a significant challenge in forging powerful graph embeddings. Even standard deep learning techniques such as Recurrent Neural Networks (RNNs) or Convolutional Neural Networks (CNNs) are not capable enough to operate on the data lying beyond 1D sequence of say words or 2D pixel-grid of images and therefore, cannot generalize to arbitrary graph structure. Recently, Graph Neural Networks (GNNs) have been proposed to alleviate such limitations but the current state is far from being mature in both theory and applications. To that end, this thesis aims at developing powerful graph embedding models for solving wide-variety of real-world problems on the graph. We study some of the major approaches for devising graph embedding namely Graph Kernel Or Spectrum and GNN. We expose and tackle some of their fundamental weakness and contribute several novel state-of-the-art graph embedding models. These models can achieve superior performance in solving many real-world problems on graphs such as node classification, graph classification or link prediction over existing methods and that too comes with desirable theoretical guarantees. We first study the capabilities of Graph Kernel or Spectrum approaches toward yielding powerful graph embeddings in terms of uniqueness, stability, sparsity and computationally efficiency. Second, we propose Graph Capsule Neural Network that can yield powerful graph embeddings by capturing much more information encoded in the graph structure in comparison with existing GNNs. Third, we devise a first ever universal and transferable GNN and thus, makes transfer learning possible in graph domain. Specifically with this particular GNN, graph embeddings can be shared and transfered across different models and domains, reaping the huge benefits of transfer learning. Lastly, there is a dearth of theoretical explorations of GNN models such as their generalization properties. We take the first step towards developing a deeper theoretical understanding of GNN models by analyzing their stability and deriving their generalization guarantees. To the best of our knowledge, we are the first to study stability bounds on graph learning in a semi-supervised setting and derive related generalization bounds for GNN models. In summary, this thesis contributes several state-of-the-art graph embeddings and novel graph theory, specifically (i) Powerful Graph Embedding called Family of Graph Spectral Distances (Fgsd) (ii) Highly Informative GNN Called Graph Capsule Neural Network (GCAPS) (iii) Universal and Transferable GNN called Deep Universal and Transferable Graph Neural Network (DUGNN) (iv) Stability Theory and Generalization Guarantees of GNN.
  • No Thumbnail Available
    Item
    Weaving Through Neural Webs: Measurement of How Students Connect the Visual Literacy Skill of Horizontally Translating Across a Chemistry Curriculum
    (2022-12)
    Biochemistry is an upper-division course that teaches topics using visual representations of systems of data, which can be a challenging way for students to learn. To lessen their cognitive load, students may find that improving visual literacy skills aids their understanding of biochemistry. There is little existing research that assesses how students interpret and store biochemical information and representations in their long-term memory. Previous studies performed with undergraduate general chemistry I students measured structural knowledge, or neural networks, of topics by asking students to assign the relatedness of chemistry-related words/phrases. Our study intends to analyze neural networks of a biochemistry visual literacy skill where undergraduate chemistry and biochemistry students to rank the relatedness of biochemical representations as an alternative to words/phrases. Specifically, this study assesses students' structural knowledge of the horizontal translation visual literacy skill, relating to the oxygen binding concept in comparison to the enzyme-substrate concept. We want to determine if there are pedagogical strategies and/or course instructional modalities that impact students' neural network development toward expert-like organization of the horizontal translation visual literacy skills. Preliminary analyses assess whether students are becoming more expert-like in correlation with their exposure to chemistry and biochemistry concepts. The data from student responses are analyzed in Pathfinder against an expert reference network to generate average degree and eccentricity values, path length correlation (PLC), and neighborhood similarity (NS) values, as well as patterns in organization/chunking. Degree values indicate the most branched nodes, while eccentricity values indicate the most central node in the neural network. PLC indicates how well each node is connected, and NS values are similarities in the grouping of concepts around the central node. Moreover, patterns in organization/chunking allow for similar groupings of nodes to be assessed. Through this study, we hope to improve curricular materials for biochemistry learning, in hopes that students will become more expert-like throughout their chemistry and biochemistry sequence. The outcome of this analysis may aid the improvement of curricular materials for optimal learning and retention of information.