-
DeepLIFT: Learning Important Features Through Propagating Activation Differences
DeepLIFT is a method for assigning feature importance that compares a neuron's activation to its'reference', where the reference is the activation that the neuron has when the... -
Simple dataset for attribution methods
A simple dataset and models trained with known relative feature importance for controlled experiments on attribution methods. -
Boston housing dataset
The Boston housing dataset is a multivariate dataset containing information about housing prices in Boston. The dataset is used to demonstrate the proposed methods for...
