Beyond just ‘correct’: new logic rules to make data insights easier to read
Our previous work created a logic to find correct rules from positive and negative data. Now, we’ve added new logical equivalences to ‘shrink’ those rules, making them shorter, simpler, and more useful for humans.
In our previous research, we built a formal “simplification logic” to manage data that has both positive and negative attributes (like “passed exam” and “did not fail homework”). That logic was a huge step forward because it allowed us to find a correct, non-redundant set of rules from complex, mixed data.
But a rule can be “correct” and “non-redundant” and still be completely useless to a human.
In our 2025 paper published in Mathematics, we tackle this final step: making those correct rules simple enough to actually read.
🧐 The problem: ‘correct’ doesn’t mean ‘usable’
Imagine a data analysis tool tells a doctor that the following rule is 100% correct: if (patient_has_fever AND not_low_blood_pressure AND is_over_60 AND not_on_medication_X), then (prognosis_is_poor)
This rule is correct, but it’s long and complicated. What if the same knowledge could be expressed more simply? if (patient_has_fever AND is_over_60), then (prognosis_is_poor)
This second rule is not only correct, it’s an insight. Our old logic was good at finding the first rule. We needed a new logic to find the second.
💡 Our solution: new logical rules for conciseness
We didn’t just want a correct basis of rules; we wanted the shortest, most readable basis possible.
So, we went back to the drawing board and proved a new set of logical equivalences. These are new, formal rules of logic that allow us to take a long implication and “shrink” or “simplify” it into an equivalent, shorter one, without changing the knowledge it represents.
🛠️ How it works: an iterative ‘shrinking’ method
This isn’t just theory. We designed an iterative simplification method that takes a set of mixed-attribute implications and repeatedly applies our new rules.
The algorithm looks for complex rules and checks if one of our new equivalences can be used to make it shorter and simpler, while mathematically guaranteeing the meaning is preserved.
*
🚀 The results: fast, effective simplification
We ran experiments and the results are fantastic. Our method can take a set of complex (but already non-redundant) rules and significantly reduce their size and complexity.
The best part? The method is fast and iterative. We found that it provides “sufficiently good results in only one or two iterations.” This means you get most of the simplification benefit almost immediately, making it a fast and practical tool for any data-sensitive experiment.
🔬 Why does this matter?
This work is about making automated reasoning and AI truly interpretable.
For an expert (like a doctor, a scientist, or an educator) to trust an AI-driven discovery, they must be able to understand the rules the system has learned. A long, convoluted rule is just a “black box” with extra steps. A short, simple rule is a genuine piece of knowledge.
This method bridges the final gap between complex, correct data and simple, human-usable insight.
📖 The full paper
For the complete theoretical breakdown of the new logical equivalences and the experimental results, you can read the original open-access article in Mathematics.
New simplification rules for databases with positive and negative attributes. Authors: Domingo López-Rodríguez, Manuel Ojeda-Hernández, Carlos Bejines. Journal: Mathematics (vol. 13, issue 2, 309)