Adaptation Space Reduction Using an Explainable Framework
Alhassan Boner Diallo, Hiroyuki Nakagawa, Tatsuhiro Tsuchiya · 2021
Self-adaptive systems can decide autonomously to adapt their settings depending on the current situation of their operating environment. To increase their dependability in a dynamic environment, different techniques like evolutionary al-gorithms, artificial intelligence techniques, etc., have been widely used. Recently, machine learning has been leveraged to solve issues like discovering new knowledge at runtime or helping to cope with uncertainty. The adaptation space reduction problem and the interactions with human-in-the-loop problem are among the issues facing self-adaptive systems. In our work we propose a mechanism that can solve the former while contributing to a solution for the latter. The mechanism uses a deep learning approach that leverages explainable AI (XAI) in the process of the learning and predictions. Our approach uses a convolutional neural network (CNN) to implement the deep learning approach and the integrated gradients technique for the explainable AI (XAI). XAI helps to build trust in the system by explaining the predictions and the behavior of the deep learning model. We evaluated our approach on the MAPE-K framework of two simulated Internet of Things systems.