Using Augmented Reality Interfaces for Artificial Intelligence Systems

Abstract

Augmented reality interfaces offer users an effective environment. In this study, a visualization approach with 3D augmented reality interfaces was introduced to enable users to understand and analyze complex deep learning models in a short time. It has been investigated whether the immersive experience that augmented reality creates on the user in other systems has the same effect when analyzing these models. Two-dimensional studies on deep learning models were examined and what could be done in three dimensions was emphasized. By adding another dimension with augmented reality interfaces, a threedimensional experience is offered to the user and the results are observed. A CNN model is visualized in the application. When test data was given to the model, the feature maps, filters and connections in the layers were displayed. The application was first run in 2 dimensions, then as a desktop application, and then in 3 dimensions, on Microsoft Hololens-2, a mixed reality headset. Tasks are given to users. Usability was measured with a test called the SUM model, which included completion or non-completion situations, errors, completion times and satisfaction. Here, satisfaction was measured using ASQ(After-Scenario Questionnaire), a user satisfaction measurement questionnaire. The usability of augmented reality 3D interfaces was found to be 80%. The conclusion reached with the answers; It has been stated that users are willing to use this system, their awareness in 3 dimensions is undeniable, and these systems can be used as a feature that increases human ability in artificial intelligence systems.