We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Excessive Invariance Causes Adversarial Vulnerability

Abstract: Despite their impressive performance, deep neural networks exhibit striking failures on out-of-distribution inputs. One core idea of adversarial example research is to reveal neural network errors under such distribution shifts. We decompose these errors into two complementary sources: sensitivity and invariance. We show deep networks are not only too sensitive to task-irrelevant changes of their input, as is well-known from epsilon-adversarial examples, but are also too invariant to a wide range of task-relevant changes, thus making vast regions in input space vulnerable to adversarial attacks. We show such excessive invariance occurs across various tasks and architecture types. On MNIST and ImageNet one can manipulate the class-specific content of almost any image without changing the hidden activations. We identify an insufficiency of the standard cross-entropy loss as a reason for these failures. Further, we extend this objective based on an information-theoretic analysis so it encourages the model to consider all task-dependent features in its decision. This provides the first approach tailored explicitly to overcome excessive invariance and resulting vulnerabilities.
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
Journal reference: Proceedings of the 7th International Conference on Learning Representations (ICLR), 2019
Cite as: arXiv:1811.00401 [cs.LG]
  (or arXiv:1811.00401v4 [cs.LG] for this version)

Submission history

From: Jörn-Henrik Jacobsen [view email]
[v1] Thu, 1 Nov 2018 14:14:03 GMT (4344kb,D)
[v2] Thu, 21 Mar 2019 03:26:21 GMT (4539kb,D)
[v3] Wed, 26 Jun 2019 04:12:20 GMT (4540kb,D)
[v4] Sun, 12 Jul 2020 07:26:06 GMT (4548kb,D)

Link back to: arXiv, form interface, contact.