Explaining Deep Classification of Time-Series Data with Learned Prototypes

The emergence of deep learning networks raises a need for explainable AI so users and domain experts can be confident applying them to high-risk. In this paper, we leverage data from the latent space induced by learning models to learn stereotypical representations or "prototypes" training to elucidate the algorithmic decision-making process. We study leveraging prototypes effect classification decisions of two dimensional-series data in a few different settings: (1) electrocardiogram (ECG) to detect clinical bradycardia, a slowing of heart rate, in preterm, (2) respiration waveforms to detect apnea of prematurity, and (3) waveforms to classify spoken digits. We improve upon existing models by for increased prototype diversity and robustness, visualize how prototypes in the latent space are used by the model to distinguish, and show that prototypes are capable of learning features on two time-series data to produce explainable insights during tasks. We show that the prototypes are capable of learning-world features - bradycardia in ECG, apnea in respiration, and in speech - as well as features within sub-classes. Our novel work learned prototypical framework on two dimensional time-series data to explainable insights during classification tasks.