SMiLE: Provably Enforcing Global Relational Properties in Neural Networks

2025-11-10

AI Summary

Existing neural network formal verification methods struggle to simultaneously guarantee global relational properties—such as monotonicity, global robustness, and individual fairness—while maintaining generality and scalability. Method: This paper introduces the first full-input-space extension of the SMiLE framework, proposing a unified approach that integrates formal verification with convex optimization. By leveraging constraint propagation and relaxation techniques, it enables precise modeling of output relationships and seamless embedding of global properties. Contribution/Results: The method provides rigorous theoretical guarantees, supports diverse model architectures and task types (classification and regression), and requires no property-specific customization. Experiments demonstrate that it achieves performance comparable to specialized baselines while significantly enhancing certification strength and applicability across properties and models—offering a novel, principled, and practical pathway toward trustworthy AI.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Security and privacy of machine learning and AI applications

Abstract

Artificial Intelligence systems are increasingly deployed in settings where ensuring robustness, fairness, or domain-specific properties is essential for regulation compliance and alignment with human values. However, especially on Neural Networks, property enforcement is very challenging, and existing methods are limited to specific constraints or local properties (defined around datapoints), or fail to provide full guarantees. We tackle these limitations by extending SMiLE, a recently proposed enforcement framework for NNs, to support global relational properties (defined over the entire input space). The proposed approach scales well with model complexity, accommodates general properties and backbones, and provides full satisfaction guarantees. We evaluate SMiLE on monotonicity, global robustness, and individual fairness, on synthetic and real data, for regression and classification tasks. Our approach is competitive with property-specific baselines in terms of accuracy and runtime, and strictly superior in terms of generality and level of guarantees. Overall, our results emphasize the potential of the SMiLE framework as a platform for future research and applications.

Problem

• Enforcing global relational properties in neural networks for robustness and fairness
• Extending enforcement framework to support properties over entire input space
• Providing full satisfaction guarantees while maintaining model accuracy

Innovation

• Extends SMiLE framework for global relational properties
• Scales with model complexity and provides full guarantees
• Accommodates general properties and various network backbones