Raphaël Olivier

2 accepted papers

2021

High-Frequency Adversarial Defense for Speech and Audio

Raphaël Olivier, Bhiksha Raj, Muhammad A. Shah

ICASSP 2021accepted

Recent work suggests that adversarial examples are enabled by high-frequency components in the dataset. In the speech domain where spectrograms are used extensively, masking those components seems like a sound direction for defenses against attacks. We explore a smoothing approach based on additive…

Cited by 0 Source ↗Scholar ↗

2021

Towards Adversarial Robustness Via Compact Feature Representations

Muhammad A. Shah, Raphaël Olivier, Bhiksha Raj

ICASSP 2021accepted

Deep Neural Networks (DNNs), while providing state-of-the-art performance in a wide variety of tasks, have been shown to be vulnerable to adversarial attacks. Recent studies have posited that this vulnerability arises because DNNs operate over a grossly overspecified input space with very sparse hum…

Cited by 0 Source ↗Scholar ↗