Where Does Trust Break Down? A Quantitative Trust Analysis of Deep Neural Networks via Trust Matrix and Conditional Trust Densities

With tremendous rise in deep learning adoption comes questions about the trustworthiness of neural networks that power a variety applications. In this work, we introduce concept trust matrix, novel quantification strategy leverages recently introduced question-answer metric by Wong et al. to provide deeper, more detailed insights into where breaks down for given network set questions. More specifically, matrix defines expected actor-oracle answer scenario, allowing one quickly spot areas low needs be addressed order improve network. We further extend densities with notion conditional densities.