QPM: Discrete Optimization for Globally Interpretable Image Classification

2025-02-27

AI Summary

To address the lack of global interpretability in deep learning image classification models for safety-critical applications, this paper proposes a discrete optimization-based framework for globally interpretable modeling. The core idea is to learn a minimal, class-shared set of binary features—typically only five—each corresponding to semantically clear, cross-class comparable visual concepts. Methodologically, we formulate a discrete optimization objective integrating similarity priors and interpretability constraints, solved via quadratic programming and feature decoupling, with fine-grained semantic alignment enabling controllable concept generation. Our approach achieves, for the first time on both small- and large-scale benchmarks, truly global interpretability at state-of-the-art (SOTA) accuracy: it enables faithful reconstruction of the model’s overall decision logic while substantially outperforming existing interpretable models in predictive performance.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Optimization
• Search and Optimization: Learning to Search

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

Understanding the classifications of deep neural networks, e.g. used in safety-critical situations, is becoming increasingly important. While recent models can locally explain a single decision, to provide a faithful global explanation about an accurate model's general behavior is a more challenging open task. Towards that goal, we introduce the Quadratic Programming Enhanced Model (QPM), which learns globally interpretable class representations. QPM represents every class with a binary assignment of very few, typically 5, features, that are also assigned to other classes, ensuring easily comparable contrastive class representations. This compact binary assignment is found using discrete optimization based on predefined similarity measures and interpretability constraints. The resulting optimal assignment is used to fine-tune the diverse features, so that each of them becomes the shared general concept between the assigned classes. Extensive evaluations show that QPM delivers unprecedented global interpretability across small and large-scale datasets while setting the state of the art for the accuracy of interpretable models.

Problem

• Global interpretability of deep neural networks
• Binary assignment for class representation
• Discrete optimization for interpretable models

Innovation

• Discrete optimization for interpretability
• Binary feature assignment for classes
• Fine-tuning features for shared concepts