Corruption Depth: Analysis of DNN depth for Misclassification

Akshay Agarwal, Nalini K. Ratha, Mayank Vatsa, Richa Singh

Keywords: Classification Depth, Deep Representation, Robust Recognition

TL;DR: Identify the layers lead to robust image classification

Abstract: Many large and complex deep neural networks have been shown to provide higher accuracy. However, very little is known about the relationship between the complexity of the input data along with the type of noise and the depth needed for correct classification. Existing studies do not address the issue of common corruption adequately, especially in understanding what impact these corruptions leave on the individual part of a deep neural network. Therefore, we can safely assume that the classification (or misclassification) might be happening at a particular layer(s) of a network that accumulates to draw a final correct or incorrect prediction. In this paper, we introduce a novel concept called {\bf corruption depth}, which identifies the location of the network layer/depth until the misclassification persists. We assert that the identification of such layers will help in better design of the network by pruning certain layers in comparison to the purification of the entire network which is computationally heavy to do. Through our extensive experiments, we present a coherent study in comparison to the existing studies which are diverse in understanding the processing of examples through the network. Our approach also illustrates different philosophies of example memorization and a one-dimensional view of sample or query difficulty. We believe that the understanding of the corruption depth can \textbf{open a new dimension of model explainability}, where in place of just visualizing the attention map, the classification progress can be seen throughout the network.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning