HyDI: A hybrid Deep Learning-Inductive Logic Programming ensemble for multi-label classification

2026-09-29

AI Summary

This study addresses the limited interpretability of deep learning in multi-level multi-label classification by proposing a hybrid architecture that integrates deep learning with Inductive Logic Programming (ILP). The core innovation lies in replacing leaf-node deep learning models with ILP-derived rule classifiers, thereby enabling transparent reasoning through the synergistic fusion of visual and textual features. Applied to the ChEBI ontology dataset, the proposed method successfully generates interpretable rules for 314 classes and provides both global and local multidimensional explanations. By bridging symbolic logic with neural representations, this approach significantly enhances the transparency and trustworthiness of complex hierarchical classification systems, offering a principled solution to the black-box limitations inherent in conventional deep learning pipelines for structured prediction tasks.

Abstract

While attaining remarkable results for many applications, Deep Learning models are notoriously difficult to explain. This work introduces HyDI, a hybrid ensemble architecture for hierarchical multi-label classification. It combines a Deep Learning (DL) model with rule-based classifiers generated by Inductive Logic Programming (ILP). For leaf classes of the label hierarchy, the rule-based classifiers replace the DL model, leading to more transparent classification results. HyDI is applied to the Chemical Entities of Biological Interest (ChEBI) ontology, providing ILP-generated rules for 314 classes. For these classes, HyDI can generate global explanations as well as local explanations that combine visual and text-based descriptions.

Problem

• Explainability
• Deep Learning
• Multi-label Classification
• Hierarchical Classification

Innovation

• Hybrid Ensemble Architecture
• Inductive Logic Programming
• Hierarchical Multi-label Classification
• Explainable AI
• Deep Learning