Data-centric Prediction Explanation via Kernelized Stein Discrepancy

2024-03-22

International Conference on Learning Representations

AI Summary

Existing example-based prediction explanation methods often rely on model parameters or latent representations to measure sample similarity, resulting in coarse-grained explanations, high computational overhead, and limited model awareness. To address these limitations, we propose HD-Explain—a data-driven, model-agnostic, fine-grained explanation method. HD-Explain is the first to leverage Kernelized Stein Discrepancy (KSD) to construct a model-aware, parameterized kernel function that efficiently identifies the most supportive training samples for a given test prediction. Crucially, it requires neither gradient backpropagation nor model retraining. Empirical evaluation on multi-class classification tasks demonstrates that HD-Explain surpasses state-of-the-art baselines in explanation accuracy, consistency, and computational efficiency. Moreover, it exhibits strong simplicity, robustness, and scalability—making it broadly applicable across diverse model architectures and datasets.

Technology Category

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

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
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Existing example-based prediction explanation methods often bridge test and training data points through the model's parameters or latent representations. While these methods offer clues to the causes of model predictions, they often exhibit innate shortcomings, such as incurring significant computational overhead or producing coarse-grained explanations. This paper presents a Highly-precise and Data-centric Explan}ation (HD-Explain) prediction explanation method that exploits properties of Kernelized Stein Discrepancy (KSD). Specifically, the KSD uniquely defines a parameterized kernel function for a trained model that encodes model-dependent data correlation. By leveraging the kernel function, one can identify training samples that provide the best predictive support to a test point efficiently. We conducted thorough analyses and experiments across multiple classification domains, where we show that HD-Explain outperforms existing methods from various aspects, including 1) preciseness (fine-grained explanation), 2) consistency, and 3) computation efficiency, leading to a surprisingly simple, effective, and robust prediction explanation solution.

Problem

• Improves coarse-grained prediction explanations efficiently
• Reduces computational overhead in example-based methods
• Enhances explanation precision using Kernelized Stein Discrepancy

Innovation

• Uses Kernelized Stein Discrepancy for explanations
• Encodes model-dependent data correlation via kernel
• Efficiently identifies predictive support samples