It assigns weights based on class probability to

influence the local surrogate.

It assigns weights based on prediction

confidence to improve local accuracy.

It assigns weights based on feature importance

to improve local fidelity.

It assigns weights

based on sample distance to influence the local surrogate.