An Explainable Deep Learning Approach for Adaptation Space Reduction
Alhassan Boner Diallo, Hiroyuki Nakagawa, Tatsuhiro Tsuchiya · 2020
The problem of adaptation space reduction has been tackled recently using machine learning. While the result has been interesting, the accuracy of the predictions was not outstanding. In this paper, we propose a deep learning approach using a convolutional neural network (CNN) to tackle the adaptation space reduction. As the inherent nature of deep learning models which behave like a black box makes them difficult to follow and to interpret, we also explore the use of Explainable AI (XAI) in the process of the learning and prediction. XAI helps to build trust in the system by explaining the predictions and the behavior of the deep learning model. We plan to evaluate our approach on two simulated IoT applications for smart environment monitoring.