Understanding Deep Networks via Extremal Perturbations and Smooth Masks

TL;DR

This paper introduces extremal perturbations and smooth masks for more interpretable attribution in deep networks, removing hyper-parameters and analyzing intermediate layers to better understand model behavior.

Contribution

It proposes a theoretically grounded method for perturbation-based attribution, including new area constraints and smooth perturbations, and extends analysis to intermediate layers.

Findings

Extremal perturbations are sensitive to spatial properties of networks.

The method identifies salient channels for classification.

TorchRay library facilitates interpretability in PyTorch.

Abstract

The problem of attribution is concerned with identifying the parts of an input that are responsible for a model's output. An important family of attribution methods is based on measuring the effect of perturbations applied to the input. In this paper, we discuss some of the shortcomings of existing approaches to perturbation analysis and address them by introducing the concept of extremal perturbations, which are theoretically grounded and interpretable. We also introduce a number of technical innovations to compute extremal perturbations, including a new area constraint and a parametric family of smooth perturbations, which allow us to remove all tunable hyper-parameters from the optimization problem. We analyze the effect of perturbations as a function of their area, demonstrating excellent sensitivity to the spatial properties of the deep neural network under stimulation. We also extend perturbation analysis to the intermediate layers of a network. This application allows us to identify the salient channels necessary for classification, which, when visualized using feature inversion, can be used to elucidate model behavior. Lastly, we introduce TorchRay, an interpretability library built on PyTorch.