Evaluating the Visualization of What a Deep Neural Network Has Learned

• Published in: IEEE transaction on neural networks and learning systems Vol. 28; no. 11; pp. 2660 - 2673
• Main Authors: Samek, Wojciech, Binder, Alexander, Montavon, Gregoire, Lapuschkin, Sebastian, Muller, Klaus-Robert
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.11.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithm design and analysis; Artificial neural networks; Biological neural networks; Classification; Convolutional neural networks; Data collection; Decision theory; Deconvolution; explaining classification; Heating; Image classification; interpretable machine learning; Learning algorithms; Learning systems; Machine learning; Neural networks; Neurons; Object recognition; Pixels; relevance models; Sensitivity; Speech recognition; Task complexity; Test procedures; Visualization; Algorithm design and analysis; Learning systems; Deconvolution; Sensitivity; Heating; Neurons; Biological neural networks
• ISSN: 2162-237X; 2162-2388; 2162-2388
• DOI: 10.1109/TNNLS.2016.2599820

Deep neural networks (DNNs) have demonstrated impressive performance in complex machine learning tasks such as image classification or speech recognition. However, due to their multilayer nonlinear structure, they are not transparent, i.e., it is hard to grasp what makes them arrive at a particular classification or recognition decision, given a new unseen data sample. Recently, several approaches have been proposed enabling one to understand and interpret the reasoning embodied in a DNN for a single test image. These methods quantify the "importance" of individual pixels with respect to the classification decision and allow a visualization in terms of a heatmap in pixel/input space. While the usefulness of heatmaps can be judged subjectively by a human, an objective quality measure is missing. In this paper, we present a general methodology based on region perturbation for evaluating ordered collections of pixels such as heatmaps. We compare heatmaps computed by three different methods on the SUN397, ILSVRC2012, and MIT Places data sets. Our main result is that the recently proposed layer-wise relevance propagation algorithm qualitatively and quantitatively provides a better explanation of what made a DNN arrive at a particular classification decision than the sensitivity-based approach or the deconvolution method. We provide theoretical arguments to explain this result and discuss its practical implications. Finally, we investigate the use of heatmaps for unsupervised assessment of the neural network performance.