Do Convolutional Neural Networks Learn Class Hierarchy?

• Published in: IEEE transactions on visualization and computer graphics Vol. 24; no. 1; pp. 152 - 162
• Main Authors: Bilal, Alsallakh, Jourabloo, Amin, Mao Ye, Xiaoming Liu, Liu Ren
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Analytics; Artificial neural networks; Confusion; confusion matrix; Convolutional Neural Networks; Data visualization; deep learning; Detectors; Feature extraction; Identification methods; Image classification; Image recognition; large-scale classification; Neural networks; Neurons; Structural hierarchy; Training; Training data
• ISSN: 1077-2626; 1941-0506; 1941-0506
• DOI: 10.1109/TVCG.2017.2744683

Convolutional Neural Networks (CNNs) currently achieve state-of-the-art accuracy in image classification. With a growing number of classes, the accuracy usually drops as the possibilities of confusion increase. Interestingly, the class confusion patterns follow a hierarchical structure over the classes. We present visual-analytics methods to reveal and analyze this hierarchy of similar classes in relation with CNN-internal data. We found that this hierarchy not only dictates the confusion patterns between the classes, it furthermore dictates the learning behavior of CNNs. In particular, the early layers in these networks develop feature detectors that can separate high-level groups of classes quite well, even after a few training epochs. In contrast, the latter layers require substantially more epochs to develop specialized feature detectors that can separate individual classes. We demonstrate how these insights are key to significant improvement in accuracy by designing hierarchy-aware CNNs that accelerate model convergence and alleviate overfitting. We further demonstrate how our methods help in identifying various quality issues in the training data.