Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanations

2025-06-10

AI Summary

Existing provably faithful explanation methods provide formal guarantees but suffer from high computational overhead and poor scalability. To address this, we propose an abstraction-refinement framework: first, structurally abstract the original neural network to rapidly generate sufficient explanations; then, iteratively refine the abstract model on-demand using neural network verification techniques, ensuring that explanations remain provably valid on the original network. This work is the first to introduce model abstraction into provable explanation generation, achieving a principled trade-off between efficiency and formal soundness. The framework supports multi-granularity attribution analysis, revealing the robustness of predictions across abstraction levels. Experimental evaluation demonstrates speedups of several-fold to over an order of magnitude in computing provably sufficient explanations, while maintaining 100% verification accuracy across multiple benchmarks.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• 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
• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity

Abstract

Despite significant advancements in post-hoc explainability techniques for neural networks, many current methods rely on heuristics and do not provide formally provable guarantees over the explanations provided. Recent work has shown that it is possible to obtain explanations with formal guarantees by identifying subsets of input features that are sufficient to determine that predictions remain unchanged using neural network verification techniques. Despite the appeal of these explanations, their computation faces significant scalability challenges. In this work, we address this gap by proposing a novel abstraction-refinement technique for efficiently computing provably sufficient explanations of neural network predictions. Our method abstracts the original large neural network by constructing a substantially reduced network, where a sufficient explanation of the reduced network is also provably sufficient for the original network, hence significantly speeding up the verification process. If the explanation is in sufficient on the reduced network, we iteratively refine the network size by gradually increasing it until convergence. Our experiments demonstrate that our approach enhances the efficiency of obtaining provably sufficient explanations for neural network predictions while additionally providing a fine-grained interpretation of the network's predictions across different abstraction levels.

Problem

• Provable explanations for neural network predictions
• Scalability challenges in computing sufficient explanations
• Efficient abstraction-refinement technique for verification

Innovation

• Abstraction-refinement technique for explanations
• Reduced network speeds up verification
• Iterative refinement ensures sufficient explanations