Towards Gradient-based Time-Series Explanations through a SpatioTemporal Attention Network

However, it is not desirable to apply AI fully autonomously as wrong outcomes of AI models in high-stake domains could have serious impacts on people. Regardless of the performance of an AI model, the end-users desire to understand the evidence on the outcome of an AI model [35]. A growing body of research investigates how to generate explanations of an AI model and augment user's decision-making tasks [2, 18, 25]. Researchers have explored various techniques to make AI interpretable and explainable [15]. These explainable AI techniques can be broadly categorized into inherently interpretable models (e.g.