Improving testing of multi-agent systems: An innovative deep learning strategy for automatic, scalable, and dynamic error detection and optimisation

Abstract

In this paper, a novel method is introduced for automated, scalable, and dynamic identification of errors in various behavioural versions of a multi-agent system under test, employing deep learning techniques. It is designed to enable accurate error detection, thus opening new possibilities for improving and optimising traditional testing techniques. The approach consists of two phases. The first phase is the training of a deep learning model using randomly generated inputs and predicted outputs generated from the behavioural model of each version. The second phase consists of detecting errors in the multi-agent system under test by replacing the predicted outputs with which the model is trained with execution outputs. The envisioned strategy is put into action through a real case study, which serves to vividly showcase and affirm its practical efficacy.