LCANets++: Robust Audio Classification using Multi-layer Neural Networks with Lateral Competition
This addresses robustness issues in audio classification for applications like speech recognition, but it is incremental as it builds on existing LCANets from computer vision.
The paper tackles the problem of audio classifiers being vulnerable to perturbations and adversarial attacks, and introduces LCANets++, which extends neuro-inspired CNNs with multi-layer sparse coding to improve robustness. The result shows that LCANets++ are more robust than standard CNNs and previous LCANets against noise and attacks like FGSM, though no concrete numbers are provided.
Audio classification aims at recognizing audio signals, including speech commands or sound events. However, current audio classifiers are susceptible to perturbations and adversarial attacks. In addition, real-world audio classification tasks often suffer from limited labeled data. To help bridge these gaps, previous work developed neuro-inspired convolutional neural networks (CNNs) with sparse coding via the Locally Competitive Algorithm (LCA) in the first layer (i.e., LCANets) for computer vision. LCANets learn in a combination of supervised and unsupervised learning, reducing dependency on labeled samples. Motivated by the fact that auditory cortex is also sparse, we extend LCANets to audio recognition tasks and introduce LCANets++, which are CNNs that perform sparse coding in multiple layers via LCA. We demonstrate that LCANets++ are more robust than standard CNNs and LCANets against perturbations, e.g., background noise, as well as black-box and white-box attacks, e.g., evasion and fast gradient sign (FGSM) attacks.