SPADE: Sparsity-Guided Debugging for Deep Neural Networks

TL;DR

SPADE introduces a novel sample-specific sparsity-based preprocessing method that enhances neural network interpretability without altering the trained model or its inference behavior.

Contribution

This work is the first to incorporate sparsity directly into the interpretation process as a preprocessing step, improving interpretability methods without retraining or modifying the original network.

Findings

SPADE improves the accuracy of image saliency maps across multiple interpretability methods.

SPADE enhances the usefulness of neuron visualizations for understanding network behavior.

Preprocessing with SPADE does not affect the model's inference performance.

Abstract

It is known that sparsity can improve interpretability for deep neural networks. However, existing methods in the area either require networks that are pre-trained with sparsity constraints, or impose sparsity after the fact, altering the network's general behavior. In this paper, we demonstrate, for the first time, that sparsity can instead be incorporated into the interpretation process itself, as a sample-specific preprocessing step. Unlike previous work, this approach, which we call SPADE, does not place constraints on the trained model and does not affect its behavior during inference on the sample. Given a trained model and a target sample, SPADE uses sample-targeted pruning to provide a "trace" of the network's execution on the sample, reducing the network to the most important connections prior to computing an interpretation. We demonstrate that preprocessing with SPADE significantly increases the accuracy of image saliency maps across several interpretability methods. Additionally, SPADE improves the usefulness of neuron visualizations, aiding humans in reasoning about network behavior. Our code is available at https://github.com/IST-DASLab/SPADE.