Using Visualization of Convolutional Neural Networks in Virtual Reality for Machine Learning Newcomers

Abstract

Software systems and components are increasingly based on machine learning methods, such as Convolutional Neural Networks (CNNs). Thus, there is a growing need for common programmers and machine learning newcomers to understand the general functioning of these algorithms. However, as neural networks are complex in nature, novel presentation means are required to enable rapid access to the functionality. For that purpose, we examine how CNNs can be visualized in Virtual Reality (VR), as a virtual environment offers the opportunity to focus users on content through effects such as immersion and presence. With a first exploratory study, we confirmed that our visualization approach is both intuitive to use and conductive to learning. Moreover, users indicated an increased motivation to learning due to the virtual environment. Based on our findings, we propose a follow-up study that specifically compares the benefits of a virtual visualization approach to a traditional desktop visualization.