A Generic Machine Learning Model Using Semantic Primitives

A generic machine learning model is proposed based on the theory of semantic primitives. It contains five components. Each of the component contains a set of primitives machine learning techniques. The input component contains primitives user initiative, system initiative, and system selection. The transformation component is composed of five primitives: generalization, specialization, integration, GA operators, and analogy. The output component consists of three primitives, namely, display, revision, and plan-proposing. The control module component supports following common control paradigms: agenda-based, competition-based, constraint-based, and user-based controls. The knowledge base component provides proper initial knowledge, functions, and heuristics. This model permits the derivation of various existent machine learning paradigms through proper integration of the primitives in each of the components. A generic machine learning shell can then be constructed based on the model. It supports a set of derivation rules allowing the user to easily derive specific machine learning shells to fit various applications.