LCANets++: Robust Audio Classification using Multi-layer Neural Networks with Lateral Competition

TL;DR

LCANets++ introduces multi-layer sparse coding with lateral competition to improve robustness in audio classification, effectively resisting noise and adversarial attacks while reducing reliance on labeled data.

Contribution

The paper extends LCANets to multiple layers with sparse coding, enhancing robustness and learning efficiency in audio recognition tasks.

Findings

LCANets++ outperforms standard CNNs in robustness against perturbations.

LCANets++ demonstrates resilience to both black-box and white-box adversarial attacks.

Sparse multi-layer coding reduces dependence on labeled data.

Abstract

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.