braindecode.visualization.integrated_gradients

braindecode.visualization.integrated_gradients(model, x, target, baseline=None, steps=50)

Integrated Gradients (Sundararajan et al., 2017) via captum.

Path integral of input gradients from baseline to x evaluated by Gauss-Legendre quadrature with steps evaluations (captum’s default). Default baseline is zeros. Even moderate steps values satisfy the completeness axiom (sum of attributions ≈ f(x) - f(baseline)).

Examples using braindecode.visualization.integrated_gradients

Interpretability of EEG Decoders

Interpretability of EEG Decoders