Data-centric Prediction Explanation via Kernelized Stein Discrepancy

TL;DR

This paper addresses the need for fine-grained, efficient prediction explanations by introducing HD-Explain, a data-centric, post-hoc method that uses Kernelized Stein Discrepancy to define a model-dependent kernel encoding data correlations. By relaxing the marginal input distribution to the training data distribution and employing the score function derived from the model, it retrieves top-k training samples that provide the strongest predictive support for a test point without perturbing the model. The approach, metrics (Hit Rate, Coverage, Run Time), and experiments across CIFAR-10, SVHN, and medical imaging datasets demonstrate improved precision, consistency, and scalability over existing methods like Influence Function, RPS, and TracIn, with insights into kernel choices (Linear, RBF, IMQ) and last-layer variants. The work advances transparency in ML systems by offering a faithful, instance-level explanation mechanism that integrates model-aware data correlations into the explanation process, potentially aiding trust and debugging in high-stakes settings.

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.

Figures (11)

• Figure 1: Varying of Kernelized Stein Discrepancy given the shift of training data distribution on Two Moon dataset.
• Figure 2: Demonstration of HD-Explain on 2D Rectangular synthetic dataset. Left shows the training dataset with three classes. Middle figure shows the explanation support of training data points to a given test point (as black cross), where green shows a higher KSD kernel value. Right shows the distribution of KSD kernel values (over the training set) to the test point, where only a small number of training data points provide strong support to this prediction.
• Figure 3: Qualitative evaluation of various example-based explanation methods using CIFAR10. We show three scenarios where the target model makes a) a highly-confident prediction that matches ground truth label, b) a low-confident prediction that matches ground truth label, c) low-confident prediction that does not match ground truth label (which is a bird). For each sub plot, we show top-3 influential training data points picked by the explanation methods for the test example.
• Figure 4: Qualitative evaluation of various example-based explanation methods using SVHN. We show two scenarios where the target model makes a-b) a highly-confident prediction that matches ground truth label, c) a low-confident prediction that matches ground truth label. For each sub plot, we show top-3 influential training data points picked by the explanation methods for the test example. We include two samples of high-confidence correct predictions to show the overlap of explanations.
• Figure 5: Quantitative explanation comparison among candidate example-based explanation methods. Data augmentation strategy used is Noise Injection. Error bar shows 95% confidence interval.

Theorems & Definitions (1)

• Definition K.1: Data attribution