Induction of logic functions

Inductive learning makes a hypothesis of the general input-output rules that might underlie training examples and forecasts outputs for hitherto-unseen inputs. Decision trees and neural networks are widely used for inductive inference. This paper proposes a novel approach that applies logic minimization for inductive inference. Training examples are transformed to incompletely specified Boolean functions and then the functions are simplified. Logic minimization on the two-level sum of products Boolean function is effective for learning shallow functions. For learning deep functions, such as exclusive-or functions, multilevel logic optimization techniques such as functional decomposition are effective