TY - GEN
T1 - Adversarially training for audio classifiers
AU - Sallo, Raymel Alfonso
AU - Esmaeilpour, Mohammad
AU - Cardinal, Patrick
N1 - Publisher Copyright:
© 2020 IEEE
PY - 2020
Y1 - 2020
N2 - In this paper, we investigate the potential effect of the adversarially training on the robustness of six advanced deep neural networks against a variety of targeted and non-targeted adversarial attacks. We firstly show that, the ResNet-56 model trained on the 2D representation of the discrete wavelet transform appended with the tonnetz chromagram outperforms other models in terms of recognition accuracy. Then we demonstrate the positive impact of adversarially training on this model as well as other deep architectures against six types of attack algorithms (white and black-box) with the cost of the reduced recognition accuracy and limited adversarial perturbation. We run our experiments on two benchmarking environmental sound datasets and show that without any imposed limitations on the budget allocations for the adversary, the fooling rate of the adversarially trained models can exceed 90%. In other words, adversarial attacks exist in any scales, but they might require higher adversarial perturbations compared to non-adversarially trained models.
AB - In this paper, we investigate the potential effect of the adversarially training on the robustness of six advanced deep neural networks against a variety of targeted and non-targeted adversarial attacks. We firstly show that, the ResNet-56 model trained on the 2D representation of the discrete wavelet transform appended with the tonnetz chromagram outperforms other models in terms of recognition accuracy. Then we demonstrate the positive impact of adversarially training on this model as well as other deep architectures against six types of attack algorithms (white and black-box) with the cost of the reduced recognition accuracy and limited adversarial perturbation. We run our experiments on two benchmarking environmental sound datasets and show that without any imposed limitations on the budget allocations for the adversary, the fooling rate of the adversarially trained models can exceed 90%. In other words, adversarial attacks exist in any scales, but they might require higher adversarial perturbations compared to non-adversarially trained models.
KW - Adversarial attack
KW - Adversarially training
KW - AlexNet
KW - Chromagram
KW - Deep neural network
KW - Discrete wavelet transformation (DWT)
KW - GoogLeNet
KW - ResNet
KW - Short-Time Fourier Transformation (STFT)
KW - Sound classification
KW - Spectrogram
KW - Tonnetz features
KW - VGG
UR - https://www.scopus.com/pages/publications/85110505144
U2 - 10.1109/ICPR48806.2021.9413263
DO - 10.1109/ICPR48806.2021.9413263
M3 - Contribution to conference proceedings
AN - SCOPUS:85110505144
T3 - Proceedings - International Conference on Pattern Recognition
SP - 9569
EP - 9576
BT - Proceedings of ICPR 2020 - 25th International Conference on Pattern Recognition
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 25th International Conference on Pattern Recognition, ICPR 2020
Y2 - 10 January 2021 through 15 January 2021
ER -