Towards Modern Rule-Based Learning

Symbolic classification is a subfield of symbolic learning focused on extracting a collection of mutually exclusive logical rules for classification. This is typically achieved by learning intermediate models, such as decision trees or decision lists. In this paper, we present Modal Sequential Covering (MSC), a decision list learning algorithm that generalizes several existing proposals in the literature. We also provide an early open-source implementation of this algorithm in Julia, integrated within Sole, a comprehensive end-to-end framework for modern symbolic AI. An experimental comparison with available tools reveals that MSC allows us to learn simpler but equally performant models. The integration of MSC into Sole enables manipulating and visualizing the extracted knowledge in logical form. As Sole is designed for symbolic learning with modal and propositional logics, this work lays the foundation for further generalization to non-tabular data. © 2024 Copyright for this paper by its authors.

Symbolic classification is a subfield of symbolic learning focused on extracting a collection of mutually exclusive logical rules for classification. This is typically achieved by learning intermediate models, such as decision trees or decision lists. In this paper, we present Modal Sequential Covering (MSC), a decision list learning algorithm that generalizes several existing proposals in the literature. We also provide an early open-source implementation of this algorithm in Julia, integrated within Sole, a comprehensive end-to-end framework for modern symbolic AI. An experimental comparison with available tools reveals that MSC allows us to learn simpler but equally performant models. The integration of MSC into Sole enables manipulating and visualizing the extracted knowledge in logical form. As Sole is designed for symbolic learning with modal and propositional logics, this work lays the foundation for further generalization to non-tabular data.