We propose a new notion of 'non-linearity' of a network layer with respect to an input batch that is based on its proximity to a linear system, which is reflected in the non-negative rank of the activation matrix.
We measure this non-linearity by applying non-negative factorization to the activation matrix.
Considering batches of similar samples, we find that high non-linearity in deep layers is indicative of memorization. Furthermore, by applying our approach layer-by-layer, we find that the mechanism for memorization consists of distinct phases. We perform experiments on fully-connected and convolutional neural networks trained on several image and audio datasets. Our results demonstrate that as an indicator for memorization, our technique can be used to perform early stopping.

Detecting Memorization in ReLU Networks

"Too Long; Didn't Read": a short sentence describing your paper

Comma separated list of author email addresses, lowercased, in the same order as above. For authors with existing OpenReview accounts, please make sure that the provided email address(es) match those listed in the author's profile. Please provide real emails; identities will be anonymized.

Comma separated list of author names. Please provide real names; identities will be anonymized.

Comma separated list of keywords describing this paper

Detecting Memorization in ReLU Networks

Edo Collins, Siavash Arjomand Bigdeli, Sabine Süsstrunk