6. Confidence Assessment of AI Models in Simulated Industrial Environments

By Spyros Theodoropoulos, Department of Digital Systems, University of Piraeus, Greece and Department of Electrical and Computer Engineering, National Technical University of Athens, Greece | Dimitrios Dardanis, Department of Digital Systems, University of Piraeus, Greece | Georgios Soanidis, Department of Digital Systems, University of Piraeus, Piraeus, Greece | Jože M. Rožanec, Jožef Stefan International Postgraduate School, Slovenia and Jožef Stefan Institute, Slovenia | Panagiotis Tsanakas, Department of Electrical and Computer Engineering, National Technical University of Athens, Greece | Dimosthenis Kyriazis, Department of Digital Systems, University of Piraeus, Greece

Downloaded: 3609 times

Published: 22 Nov 2021

© 2021 Spyros Theodoropoulos | Dimitrios Dardanis | Georgios Soanidis | Jože M. Rožanec | Panagiotis Tsanakas | Dimosthenis Kyriazis

Abstract

The deployment of artificial intelligence (AI) solutions in simulated industrial environments, such as manufacturing production lines, minimizes the risks of physical damage caused by potential agent errors ormalfunctions. Leveragingsynthetic data generation and data augmentation techniques can increase the accuracy and robustness of an AI solution. To that end, artificially generated adversarial scenarios can be exploited to assess an AI agent's confidence level and quality. This chapter will present the state-of-the-art techniques that aim to increase the confi- dence assessment of manufacturing focused AI agents by spanning the fields of Reinforcement Learning, Explainable AI and Visual Analytics.