Decorrelative network architecture for robust electrocardiogram classification
2024

Improving ECG Classification with Diverse Neural Networks

publication Evidence: moderate

Author Information

Author(s): Wiedeman Christopher, Wang Ge

Primary Institution: Rensselaer Polytechnic Institute

Hypothesis

Can feature decorrelation and Fourier partitioning improve the robustness of ECG classification against adversarial attacks?

Conclusion

The proposed methods enhance the robustness of ECG classification models against adversarial attacks while maintaining performance on unperturbed data.

Supporting Evidence

  • Ensemble methods can estimate uncertainty in predictions.
  • Adversarial attacks can mislead traditional models, but diversified features improve robustness.
  • Feature decorrelation and Fourier partitioning can be applied to enhance model performance.

Takeaway

This study shows that using different features in neural networks can help them make better decisions when faced with tricky inputs, like those that try to trick them.

Methodology

The study tested ensemble methods with feature decorrelation and Fourier partitioning on ECG classification tasks against adversarial attacks.

Limitations

The study's methods may require extensive hyperparameter tuning and may not generalize to all types of adversarial attacks.

Digital Object Identifier (DOI)

10.1016/j.patter.2024.101116

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication