Abstract-- This paper presents a new application of an Exclusive-Sum-Of-Products (ESOP) minimizer EXORCISM-MV-2: to Machine Learning, and particularly, in Pattern Theory. An analysis of various logic synthesis programs has been conducted at Wright Laboratory for machine learning applications. Creating a robust and efficient Boolean minimizer for machine learning that would minimize a decomposed function cardinality (DFC) measure of functions would help to solve practical problems in application areas that are of interest to the Pattern Theory Group - especially those problems that require strongly unspecified multiple-valued-input functions with a large number of variables. For many functions, the complexity minimization of EXORCISM-MV-2 is better than that of Espresso. For small functions, they are worse than those of the Curtis-like Decomposer. However, EXORCISM is much faster, can run on problems with more variables, and significant DFC improvements have also been found. We analyze the cases when EXORCISM is worse than Espresso and propose new improvements for strongly unspecified functions.