Interpreting the Predictions of Complex ML Models by Layer-wise Relevance Propagation

Complex nonlinear models such as deep neural network (DNNs) have become an\nimportant tool for image classification, speech recognition, natural language\nprocessing, and many other fields of application. These models however lack\ntransparency due to their complex nonlinear structure and to the complex data\ndistributions to which they typically apply. As a result, it is difficult to\nfully characterize what makes these models reach a particular decision for a\ngiven input. This lack of transparency can be a drawback, especially in the\ncontext of sensitive applications such as medical analysis or security. In this\nshort paper, we summarize a recent technique introduced by Bach et al. [1] that\nexplains predictions by decomposing the classification decision of DNN models\nin terms of input variables.\n