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

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

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: Sparse Coding Frontend for Robust Neural Networks

Abstract: Deep Neural Networks are known to be vulnerable to small, adversarially crafted, perturbations. The current most effective defense methods against these adversarial attacks are variants of adversarial training. In this paper, we introduce a radically different defense trained only on clean images: a sparse coding based frontend which significantly attenuates adversarial attacks before they reach the classifier. We evaluate our defense on CIFAR-10 dataset under a wide range of attack types (including Linf , L2, and L1 bounded attacks), demonstrating its promise as a general-purpose approach for defense.
Comments: International Conference on Learning Representations (ICLR) 2021 Workshop on Security and Safety in Machine Learning Systems
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2104.05353 [cs.LG]
  (or arXiv:2104.05353v1 [cs.LG] for this version)

Submission history

From: Can Bakiskan [view email]
[v1] Mon, 12 Apr 2021 11:14:32 GMT (1175kb,D)

Link back to: arXiv, form interface, contact.