|
V14 N2 Paper 3
|
Annals of the MS in Computer Science and Information Systems at
UNC Wilmington
|
Fall 2020
|
Visualizing the Maturation of an Artificial Neural Network
Cody Polera
Committee
Abstract
The objective of this project is to create a visualization application capable of demonstrating in three-dimensional virtual space the abstract learning processes an artificial neural network (ANN) undergoes. This application will consist of three main components: an ANN implementation, a framework to store the states of the network, and the graphical visualization that can ingest the stored states. This application will have an example implementation to leverage this visualization as an educational tool. The example tool is intended for use by post-secondary students learning about ANN to improve their understanding of the general concepts of ANN. An ANN undergoes numerous changes and adaptations to the various nodes that make up the network via the learning process. As a network learns, aspects such as the weights given to neurons in a layer change along with their values due to a complex series of formulas occurring recursively over time. As the size and scope of the network increases, it becomes more difficult to tune the performance or troubleshoot issues. By visualizing the process over time as a time-lapse, it is proposed that it would be simpler to comprehend the changes. This would be further assisted by allowing the user to explore the network in a virtual space where they can gain an even better perspective of the changes.
It was originally proposed that the project leverage virtual reality to enhance the experience, but this was ultimately removed from the project. Virtual reality capable devices are not readily available which provides a mechanical obstacle to later usabilityv testing with individuals. This obstacle is further compounded by the limitations that the global Covid-19 pandemic responses have generated. These responses limit the ability to interact with other individuals as well as provide them with devices that are handled by numerous people. The scope of the application was reduced to a desktop application which was accessible on a wide range of machines and did not require atypical devices or hardware. The scope change permitted the experiment to be safely undertaken by removing the risks associated with face-to-face interactions during the pandemic. The removal of XR also removed the risks of contamination on the devices that would be required to be shared to experience the tool in XR.
The efficacy of the tool was evaluated using a small usability study. The procedure of the usability study was having selected users download the application, answer a survey, navigate the example learning tool implementation, and then respond to a final survey. Metrics were then gathered from these survey results which indicated that while the tool was helpful overall, it is unclear if a better implementation would have improved the outcomes. Furthermore, it would be necessary to compare the results between this tool and more traditional learning mediums through further studies before any other conclusions could be drawn.
download
(pdf)
Recommended Citation:
Polera, C., Pence, T, Cummings, J., Ebrahimi, E. (2020) Visualizing the Maturation of an Artificial Neural Network. Annals of the Master of Science in Computer Science and Information Systems at UNC Wilmington, 14(2) paper 3. http://csbapp.uncw.edu/data/mscsis/full.aspx.
|
V14 N2 Paper 3
|
Annals of the MS in Computer Science and Information Systems at
UNC Wilmington
|
Fall 2020
|
