Derivative-Free Diffusion Manifold-Constrained Gradient for Unified XAI

TL;DR

This paper introduces FreeMCG, a derivative-free, manifold-constrained gradient approximation method that improves explainability of neural networks by requiring only output access and producing more faithful, human-aligned attributions.

Contribution

The paper presents FreeMCG, a novel derivative-free approach using ensemble Kalman filters and diffusion models for better model explanations on the data manifold.

Findings

Achieves state-of-the-art results in counterfactual generation.

Provides more faithful and human-aligned feature attributions.

Requires only model output access, enhancing black-box applicability.

Abstract

Gradient-based methods are a prototypical family of explainability techniques, especially for image-based models. Nonetheless, they have several shortcomings in that they (1) require white-box access to models, (2) are vulnerable to adversarial attacks, and (3) produce attributions that lie off the image manifold, leading to explanations that are not actually faithful to the model and do not align well with human perception. To overcome these challenges, we introduce Derivative-Free Diffusion Manifold-Constrainted Gradients (FreeMCG), a novel method that serves as an improved basis for explainability of a given neural network than the traditional gradient. Specifically, by leveraging ensemble Kalman filters and diffusion models, we derive a derivative-free approximation of the model's gradient projected onto the data manifold, requiring access only to the model's outputs. We demonstrate the effectiveness of FreeMCG by applying it to both counterfactual generation and feature attribution, which have traditionally been treated as distinct tasks. Through comprehensive evaluation on both tasks, counterfactual explanation and feature attribution, we show that our method yields state-of-the-art results while preserving the essential properties expected of XAI tools.