RANCER: Non-Axis Aligned Anisotropic Certification with Randomized Smoothing

• Published in: 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) pp. 4661 - 4669
• Main Authors: Rumezhak, Taras, Eiras, Francisco Girbal, S. Torr, Philip H., Bibi, Adel
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.01.2023
• Subjects: adversarial attack and defense methods; Algorithms: Adversarial learning; Certification; Computer vision; Deep learning; Neural networks; Optimization; Smoothing methods
• ISSN: 2642-9381
• DOI: 10.1109/WACV56688.2023.00465

As modern networks have been proven to be unprotected from adversarial attacks and are applied in safety-critical applications, defense against them is very crucial. Many works were dedicated to this topic, but randomized smoothing has been recently proven to be an effective approach for the certified defense of deep neural networks and getting robust classifiers. Some prior results were obtained utilizing the techniques of adding extra parameters to extend the limits of the certification regions. In this way, sample-wise optimization was proposed to maximize the certification radius per input. The idea was further extended with the generalized anisotropic counterparts of ℓ 1 and ℓ 2 certificates which allow achieving larger certified region volume avoiding worst-case certification near potentially larger safe regions. However, anisotropic certification is limited by the aligned axis lacking the freedom to extend in any direction. To mitigate this constraint, in this work, we (i) revisit the anisotropic certification, provide an analysis of its non-axis aligned counterpart and propose its rotation-free extension, (ii) conduct experiments on the CIFAR-10 dataset to report the improved performance.