New Method Uses Boolean Logic and Quantum Acceleration for More Interpretable AI
-
Introduces expressive Boolean formulas as an interpretable ML model for binary classification that can provide explanations for decisions.
-
Defines the problem of training such a model as a combinatorial optimization problem using a native local search.
-
Explains how non-local moves, solved as ILP/QUBO problems, could accelerate training using a quantum or classical hardware accelerator.
-
Benchmarks the local search with and without non-local moves on real datasets, showing improved balanced accuracy with non-local moves.
-
Announces the open-source release of BoolXAI package for training classifiers with native local optimization of expressive Boolean formulas.