Network dissection: quantifying interpretability of deep visual representations
Author(s)Bau, David; Zhou, Bolei; Khosla, Aditya; Oliva, Aude; Torralba, Antonio
MetadataShow full item record
We propose a general framework called Network Dissection for quantifying the interpretability of latent representations of CNNs by evaluating the alignment between individual hidden units and a set of semantic concepts. Given any CNN model, the proposed method draws on a broad data set of visual concepts to score the semantics of hidden units at each intermediate convolutional layer. The units with semantics are given labels across a range of objects, parts, scenes, textures, materials, and colors. We use the proposed method to test the hypothesis that interpretability of units is equivalent to random linear combinations of units, then we apply our method to compare the latent representations of various networks when trained to solve different supervised and self-supervised training tasks. We further analyze the effect of training iterations, compare networks trained with different initializations, examine the impact of network depth and width, and measure the effect of dropout and batch normalization on the interpretability of deep visual representations. We demonstrate that the proposed method can shed light on characteristics of CNN models and training methods that go beyond measurements of their discriminative power. ©2017 Paper presented at the 30th IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2017), July 21-26, 2017, Honolulu, Hawaii.
DepartmentMassachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory
Proceedings, 30th IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2017)
Bau, David, Bolei Zhou, Aditya Khosla, Aude Oliva, and Antonio Torralba, "Network dissection: quantifying interpretability of deep visual representations." Proceedings, 30th IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2017) (Piscataway, N.J.: IEEE, 2017): p. 3319-27 doi 10.1109/cvpr.2017.354 ©2017 Author(s)
Author's final manuscript