On using logic synthesis for supervised classification learning
Jeffery A. Goldman, Mark L. Axtell · 2002
Learning from data is the central theme of knowledge discovery in databases (KDD) and the machine learning (ML) community. In order to handle large databases, certain assumptions are necessary to make the problem tractable. Without introducing explicit domain knowledge, a natural assumption is Occam's Razor. However, the requirement to find solutions of low complexity is not limited to KDD and ML. For example, in the logic synthesis community, low-complexity solutions are sought for realizing circuits. Although the logic synthesis paradigms discussed are certainly not new, it is still a relatively unknown phenomenon when referring to these tools' ability as machine learning programs. We demonstrate the applicability of circuit design tools to the KDD and ML communities. Specifically, we exhibit results from C4.5 (a typical machine learning algorithm), Espresso (a 2-level minimization circuit design tool),and Function Extrapolation by Recomposing Decompositions (FERD).