Timezone: »

 
Towards Achieving Adversarial Robustness Beyond Perceptual Limits
Sravanti Addepalli · Samyak Jain · Gaurang Sriramanan · Shivangi Khare · Venkatesh Babu Radhakrishnan

@
in A Blessing in Disguise: The Prospects and Perils of Adversarial Machine Learning »
The vulnerability of Deep Neural Networks to Adversarial Attacks has fuelled research towards building robust models. While most existing Adversarial Training algorithms aim towards defending against imperceptible attacks, real-world adversaries are not limited by such constraints. In this work, we aim to achieve adversarial robustness at larger epsilon bounds. We first discuss the ideal goals of an adversarial defense algorithm beyond perceptual limits, and further highlight the shortcomings of naively extending existing training algorithms to higher perturbation bounds. In order to overcome these shortcomings, we propose a novel defense, Oracle-Aligned Adversarial Training (OA-AT), that attempts to align the predictions of the network with that of an Oracle during adversarial training. The proposed approach achieves state-of-the-art performance at large epsilon bounds ($\ell_\infty$ bound of $16/255$) while outperforming adversarial training algorithms such as AWP, TRADES and PGD-AT at standard perturbation bounds ($\ell_\infty$ bound of $8/255$) as well.

Author Information

Sravanti Addepalli (Indian Institute of Science)
Samyak Jain (Indian Institute of Technology (BHU), Varanasi)
Gaurang Sriramanan (Indian Institute of Science, Bangalore)
Shivangi Khare (Indian Institute of Science)
Venkatesh Babu Radhakrishnan (Indian Institute of Science)

More from the Same Authors