Neural Logic Networks for Interpretable Classification

2025-08-11

AI Summary

Conventional neural networks achieve strong classification performance but lack interpretability, hindering verification and debugging. Method: We propose a Neuro-Logical Network (NLN), the first neural architecture to explicitly model AND, OR, and NOT logical operations; it incorporates a bias mechanism for unobserved variables to enhance rule robustness and employs a factorized IF-THEN rule structure to support compositional concept learning and logical reasoning. The framework integrates symbolic logic with probabilistic modeling and uses an improved end-to-end learning algorithm. Contribution/Results: On Boolean network discovery, NLN achieves state-of-the-art (SOTA) performance. Applied to real-world clinical data, it automatically extracts highly relevant, semantically transparent diagnostic rules—demonstrating superior model trustworthiness and clinical utility compared to black-box alternatives.

Technology Category

• Machine Learning: Neuro-Symbolic Learning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Natural Language Processing: Safety and Robustness

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Traditional neural networks have an impressive classification performance, but what they learn cannot be inspected, verified or extracted. Neural Logic Networks on the other hand have an interpretable structure that enables them to learn a logical mechanism relating the inputs and outputs with AND and OR operations. We generalize these networks with NOT operations and biases that take into account unobserved data and develop a rigorous logical and probabilistic modeling in terms of concept combinations to motivate their use. We also propose a novel factorized IF-THEN rule structure for the model as well as a modified learning algorithm. Our method improves the state-of-the-art in Boolean networks discovery and is able to learn relevant, interpretable rules in tabular classification, notably on an example from the medical field where interpretability has tangible value.

Problem

• Develop interpretable neural logic networks for classification
• Generalize networks with NOT operations and bias handling
• Propose factorized IF-THEN rules for medical interpretability

Innovation

• Generalized Neural Logic Networks with NOT operations
• Factorized IF-THEN rule structure for interpretability
• Modified learning algorithm for Boolean networks