Logical preprocessing shrinks hypothesis space for ILP
Researchers introduce a method that logically preprocesses background knowledge to shrink the hypothesis space before inductive logic programming (ILP) systems search it. The approach identifies rules that cannot be part of an optimal hypothesis regardless of training examples—such as 'even numbers cannot be odd'—and removes violating rules. Implemented using answer set programming, it was tested on visual reasoning and game playing domains, substantially reducing learning times while maintaining predictive accuracy. The work is detailed in arXiv:2506.06739.
Key facts
- Inductive logic programming (ILP) is a form of logical machine learning.
- The goal is to search a hypothesis space for a hypothesis that generalises training examples and background knowledge.
- The new approach shrinks the hypothesis space before an ILP system searches it.
- It uses background knowledge to find rules that cannot be in an optimal hypothesis regardless of training examples.
- Example: 'even numbers cannot be odd' and 'prime numbers greater than 2 are odd'.
- Implementation uses answer set programming.
- Experiments on visual reasoning and game playing domains show substantial reduction in learning times.
- Predictive accuracies are maintained.
- Paper ID: arXiv:2506.06739.
Entities
—