Automatic Generation of Work Support Behavior with Smart Glasses based on the Deep Neural Network Corresponding to Encrypted Training Data

In Japan, which has a labor shortage problem, Smart Glasses can be a powerful tool for realizing work support. However, depending on the industry field, the types of work to be supported are enormous, and designing each one is a heavy burden for the designer. Therefore, in this research, we have investigated a method for automatically generating the support behavior of the system. Here, the method by using deep learning technology that enables end-to-end learning has been proposed so that the support behavior of the system can be obtained without specialized knowledge. However, in order to use deep learning technology, it is necessary to prepare an expensive computer environment. On the other hand, many companies have started services that provide deep learning environments that can be used in the cloud. When these services are used, training data must be sent to cloud. There are cases when this action may lead to an information leak and, from this perspective, many companies are resistant to using the cloud. Therefore, in this paper, we verify the effectiveness of the learning data encryption process for the proposed support behavior modeling method. Here, the pixel-based image encryption method proposed in the previous research is applied. As a result of the experiment, it was confirmed that the obtained support behavior model has same model structure and support behavior regardless of the presence or absence of encryption processing.