Representer Point Selection for Explaining Deep Neural Networks | Chih-Kuan Yeh

We propose to explain the predictions of a deep neural network, by pointing to the set of what we call representer points in the training set, for a given test point prediction.