Derivative-Free Diffusion Manifold-Constrained Gradient for Unified XAI

2024-11-22

arXiv.org

AI Summary

Gradient-based explainable AI (XAI) methods suffer from three key limitations: dependence on white-box model access, vulnerability to adversarial perturbations, and attribution drift away from the underlying data manifold. To address these, this paper proposes a derivative-free, manifold-constrained gradient approximation framework. Our method innovatively integrates diffusion models with ensemble Kalman filtering (EnKF) to estimate gradients in a black-box setting—driven solely by model outputs and constrained to the intrinsic data manifold. Manifold projection ensures geometric consistency during optimization. Notably, our framework unifies counterfactual generation and feature attribution within a single architecture, jointly optimizing for explanation faithfulness and perceptual plausibility. Extensive evaluations across multiple benchmarks demonstrate state-of-the-art performance on both tasks, with significant improvements in robustness, model fidelity, and perceptual alignment.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Adversarial Learning & Robustness

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

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.

Problem

• Overcoming gradient-based XAI limitations in model access
• Addressing adversarial vulnerability in explainability techniques
• Ensuring explanations align with human perception and data manifold

Innovation

• Derivative-free gradient approximation using ensemble Kalman filters
• Diffusion models for manifold-constrained projections
• Unified XAI for counterfactuals and feature attribution