37th Annual Student Research Conference
Successes and Difficulties in Generating and Implementing Convolutional Neural Network Models to Detect American Sign Language Numbers
Justin A. Caringal
Dr. Ruthie Halma, Faculty Mentor
With the prevalence of AI ever-growing, neural networks can be used to solve a variety of problems both large and small. One of these cases is the real-time translation of American Sign Language into language comprehensible by a computer. This problem lacks much in-depth research and allows for the use of open-source packages and resources to be utilized to their fullest extent. This paper discusses the research and generation of an ASL convolutional neural network model with a validation accuracy and loss of 96% and 21.2% and a testing accuracy and loss of 75.2% and 81.5%. While the ASL model is not currently able to be implemented due to limitations in technology and the time constraints of the project, the project sheds light on the development of contemporary large-scale models presently in the current popular zeitgeist and provides a launch-point for future projects and improvements; future directions are discussed.
Keywords: Artificial Intelligence, Neural Networks, Convolution, Datasets, Deep Q-Learning, Computer Vision, American Sign Language
Topic(s):Computer Science
Presentation Type: Oral Presentation
Session: 108-3
Location: MG 1098
Time: 9:45