A Survey on Neural Network Interpretability

Along with the great success of deep neural networks, there is also growing concern about their black-box nature. The interpretability issue affects people's trust on learning systems. It related to many ethical problems, e.g., algorithmic discrimination. Moreover, a desired property for networks become powerful tools in other research fields, drug discovery and genomics. In this survey, we conduct comprehensive review network research. We first clarify definition as it has been used different contexts. Then elaborate importance propose novel taxonomy organized along three dimensions: type engagement (passive vs. active interpretation approaches), explanation, focus (from local global interpretability). This provides meaningful 3D view distribution papers from relevant literature two dimensions are not simply categorical but allow ordinal subcategories. Finally, summarize existing evaluation methods suggest possible directions inspired by our new taxonomy.