Interactive Visualization of Deep Learning Models in an Immersive Environment
Hikaru Nagasaka, Motoya Izuhara · 2021
The development of deep learning (DL) models has been prevalent among software engineers. However, it is difficult for non-experts to analyze and understand their behavior. Hence, we propose an interactive visualization system of DL models in an immersive environment. Because an immersive environment offers unlimited displays and visualization of high-dimensional data, it enables a comprehensive analysis on data propagations through the layers, and compares the multiple performance metrics. In this research, we implemented a prototype system, demonstrated it to machine learning engineers, and discussed the future benefits of visualizing DL models in an immersive environment. Accordingly, our concept received positive feedback; however, we inferred that most of the engineers consider the visualization technology as a unique introduction to the immersive environment.