Explaining Deep Learning Models using Causal Inference

TL;DR

This paper introduces a causal inference-based framework to interpret CNN models by constructing Structural Causal Models and ranking filters based on their counterfactual importance, enhancing understanding of deep learning models.

Contribution

It proposes a novel causal inference approach to explain CNNs, including a method to rank filters by importance using counterfactual analysis.

Findings

Framework applied to LeNet5, VGG19, ResNet32

Filters ranked by counterfactual importance

Provides a principled way to interpret CNNs

Abstract

Although deep learning models have been successfully applied to a variety of tasks, due to the millions of parameters, they are becoming increasingly opaque and complex. In order to establish trust for their widespread commercial use, it is important to formalize a principled framework to reason over these models. In this work, we use ideas from causal inference to describe a general framework to reason over CNN models. Specifically, we build a Structural Causal Model (SCM) as an abstraction over a specific aspect of the CNN. We also formulate a method to quantitatively rank the filters of a convolution layer according to their counterfactual importance. We illustrate our approach with popular CNN architectures such as LeNet5, VGG19, and ResNet32.