TY - GEN
T1 - Symbolic Learning and Generalisation of Boolean Logic Minimisation via Ternary Weight Inference
AU - Cadenas, Oswaldo
PY - 2026/4/16
Y1 - 2026/4/16
N2 - We present a novel, interpretable approach for Boolean logic minimisation based on symbolic pattern recognition and discrete optimisation with ternary weights. Our framework constructs a compact set of simplifiable logic bases, discovered through symmetry, Hamming distance heuristics, and canonical truth table analysis. We train these bases using a guided ternary update algorithm that adaptively flips weights to minimise error on Boolean expressions represented in multi-hot symbolic form. The resulting ternary weight matrix W can simplify any Boolean function through a lightweight inference process based on symbolic decomposition. Our method achieves 100% accurate minimisation for all 3-, 4-, 5-, and 6-variable logic functions across exhaustive and large-scale random validation datasets, using no gradient descent, SAT solvers, or external heuristics. This work demonstrates that symbolic Boolean generalisation is not only feasible but also efficient, interpretable, and extensible.
AB - We present a novel, interpretable approach for Boolean logic minimisation based on symbolic pattern recognition and discrete optimisation with ternary weights. Our framework constructs a compact set of simplifiable logic bases, discovered through symmetry, Hamming distance heuristics, and canonical truth table analysis. We train these bases using a guided ternary update algorithm that adaptively flips weights to minimise error on Boolean expressions represented in multi-hot symbolic form. The resulting ternary weight matrix W can simplify any Boolean function through a lightweight inference process based on symbolic decomposition. Our method achieves 100% accurate minimisation for all 3-, 4-, 5-, and 6-variable logic functions across exhaustive and large-scale random validation datasets, using no gradient descent, SAT solvers, or external heuristics. This work demonstrates that symbolic Boolean generalisation is not only feasible but also efficient, interpretable, and extensible.
KW - minimisation, synthesis
U2 - 10.1109/ACDSA67686.2026.11467915
DO - 10.1109/ACDSA67686.2026.11467915
M3 - Conference contribution
SN - 979-8-3315-7192-4
T3 - International Conference on Artificial Intelligence, Computer, Data Sciences, and Applications, ACDSA 2026
BT - International Conference on Artificial Intelligence, Computer, Data Sciences, and Applications, ACDSA 2026
PB - Institute of Electrical and Electronics Engineers Inc.
CY - Piscataway, NJ, USA
ER -