2021/11/14 by Vadim Borisov, Johannes Meier, Borisov, Vadim +9
Computer Science · #Explainable Artificial Intelligence (XAI) #Machine Learning and Data Classification #Advanced Neural Network Applications
paper · pdf · doi:10.48550/arxiv.2111.07379
Understanding the results of deep neural networks is an essential step\ntowards wider acceptance of deep learning algorithms. Many approaches address\nthe issue of interpreting artificial neural networks, but often provide\ndivergent explanations. Moreover, different hyperparameters of an explanatory\nmethod can lead to conflicting interpretations. In this paper, we propose a\ntechnique for aggregating the feature attributions of different explanatory\nalgorithms using Restricted Boltzmann Machines (RBMs) to achieve a more\nreliable and robust interpretation of deep neural networks. Several challenging\nexperiments on real-world datasets show that the proposed RBM method\noutperforms popular feature attribution methods and basic ensemble techniques.\n