Analysis and Extensions of Adversarial Training for Video Classification. (arXiv:2206.07953v1 [cs.CV])

Web: http://arxiv.org/abs/2206.07953

Adversarial training (AT) is a simple yet effective defense against
adversarial attacks to image classification systems, which is based on
augmenting the training set with attacks that maximize the loss. However, the
effectiveness of AT as a defense for video classification has not been
thoroughly studied. Our first contribution is to show that generating optimal
attacks for video requires carefully tuning the attack parameters, especially
the step size. Notably, we show that the optimal step size varies linearly with
the …

analysis arxiv classification cv training video