Multi-task Learning for Dependability Assessment of Industrial Wireless Communication Systems

Wireless communication systems for the industrial domain will rapidly grow in the forthcoming years, which is bringing a strong need for quantitative dependability assessment. Empirical formulas and artificial intelligence have been applied in dependability assessment for years. However, these methods always focused on a single task, while multiple tasks are common and have inner relationships, such as dependability scoring and prediction. These relationships are useful and can improve every task performance. Hence, we propose the multi-task learning model to solve three tasks: dependability scoring, prediction, and scoring prediction, where we design a denoising sequence to sequence neural network with a decay attention. To validate the novelty, we measured and collected a dataset from a realistic system, and then we compared the multi-task learning results of three tasks with benchmarks, which indicates superiority.