Efficient and Accurate Explanation Estimation with Distribution Compression

2024-06-26

arXiv.org

AI Summary

To address the approximation error and high computational cost arising from i.i.d. sampling in post-hoc explanation methods, this paper proposes the “Compression-then-Explanation” (CTE) paradigm. CTE establishes, for the first time, a theoretical connection between feature attribution estimation and data distribution compression; it employs kernel thinning to achieve efficient, structure-preserving compression of the underlying data distribution, thereby drastically reducing the required sample size. CTE is plug-and-play compatible with mainstream attribution methods—including SHAP and Integrated Gradients—maintaining high explanation accuracy while reducing the number of model evaluations by 2–3× (at equivalent explanation error) and accelerating inference by 2–3×, with negligible computational overhead. Moreover, CTE enhances the stability and robustness of attribution results against input perturbations and sampling variability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Interpretability, Explainability, and Transparency
• Data Mining & Knowledge Management: Data Compression

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We discover a theoretical connection between explanation estimation and distribution compression that significantly improves the approximation of feature attributions, importance, and effects. While the exact computation of various machine learning explanations requires numerous model inferences and becomes impractical, the computational cost of approximation increases with an ever-increasing size of data and model parameters. We show that the standard i.i.d. sampling used in a broad spectrum of algorithms for post-hoc explanation leads to an approximation error worthy of improvement. To this end, we introduce Compress Then Explain (CTE), a new paradigm of sample-efficient explainability. It relies on distribution compression through kernel thinning to obtain a data sample that best approximates its marginal distribution. CTE significantly improves the accuracy and stability of explanation estimation with negligible computational overhead. It often achieves an on-par explanation approximation error 2-3x faster by using fewer samples, i.e. requiring 2-3x fewer model evaluations. CTE is a simple, yet powerful, plug-in for any explanation method that now relies on i.i.d. sampling.

Problem

• Machine Learning Model Interpretability
• Computational Cost Reduction
• Feature Importance Estimation

Innovation

• CTE Method
• Data Compression
• Feature Importance Estimation