Explainable AI using expressive Boolean formulas

The article discusses explainable AI using expressive Boolean formulas, which are interpretable machine learning models for binary classification problems. Expressive Boolean formulas consist of literals (variables or negated variables) and operators (like AND, OR, etc.) that operate on the literals.

Specifically, the article covers:

• Motivations for explainable and interpretable AI models, especially in high-stakes decision-making scenarios
• Definition and advantages of expressive Boolean formulas as an interpretable ML model
• The problem of training expressive Boolean formulas by solving a combinatorial optimization problem
• A native local solver for training the models using local moves
• Using non-local moves, proposed by solving ILPs/QUBOs on quantum computers, to potentially accelerate training
• Benchmark results showing benefits of non-local moves for higher model complexities
• Introduction to an open-source BoolXAI package for training expressive Boolean formula models