Not Just a Black Box: Learning Important Features Through Propagating Activation Differences
It addresses the need for interpretability in neural networks for applications where understanding model decisions is essential, representing an incremental improvement over existing methods.
The paper tackles the interpretability problem of neural networks by introducing DeepLIFT, a method that computes importance scores by comparing neuron activations to reference activations, showing significant advantages over gradient-based methods in applications like natural images and genomic data.
Note: This paper describes an older version of DeepLIFT. See https://arxiv.org/abs/1704.02685 for the newer version. Original abstract follows: The purported "black box" nature of neural networks is a barrier to adoption in applications where interpretability is essential. Here we present DeepLIFT (Learning Important FeaTures), an efficient and effective method for computing importance scores in a neural network. DeepLIFT compares the activation of each neuron to its 'reference activation' and assigns contribution scores according to the difference. We apply DeepLIFT to models trained on natural images and genomic data, and show significant advantages over gradient-based methods.