Baseline Computation for Attribution Methods Based on Interpolated Inputs

TL;DR

This paper proposes a method to compute effective baselines for input attribution techniques in neural networks, introducing a novel RSI-Grad-CAM method and demonstrating its effectiveness.

Contribution

It introduces a new baseline computation approach and a novel RSI-Grad-CAM attribution method for neural network interpretability.

Findings

Effective baseline computation improves attribution quality

RSI-Grad-CAM provides more accurate attributions

Method demonstrates robustness across different models

Abstract

We discuss a way to find a well behaved baseline for attribution methods that work by feeding a neural network with a sequence of interpolated inputs between two given inputs. Then, we test it with our novel Riemann-Stieltjes Integrated Gradient-weighted Class Activation Mapping (RSI-Grad-CAM) attribution method.