An End-to-End Trainable Deep Convolutional Neuro-Fuzzy Classifier

A key challenge in artificial intelligence is the well-known tradeoff between the interpretability of an algorithm, and its accuracy. Designing interpretable, highly accurate AI models is considered essential to broad acceptance of AI technology, and is the focus of the eXplainable Artificial Intelligence (XAI) community. We report on the design of a new deep neural network that achieves improved interpretability without sacrificing accuracy. Our design is a hybrid deep learning algorithm based in part upon fuzzy logic, which performs as accurately as existing convolutional neural networks. The network is an end-to-end trainable deep convolutional network, which replaces the final dense layers (the classifier component) with a modified ANFIS. We exploit the transparency of fuzzy logic by deriving explanations, in the form of saliency maps, based on the fuzzy rules learned in the ANFIS component.