Structured Machine Learning for Data Analytics and Modeling: Intelligent Security as an Example

摘要

Structured machine learning refers to learning a structured hypothesis from data with rich internal structure. We apply semantics-enabled (semi-)supervised learning for perfect and imperfect domain knowledge to fulfill the vision of structured machine learning for big data analytics and modeling. First, domain knowledge is modeled as RDF(S) ontologies, and SPARQL enables approximate queries for a type-labeled training dataset from ontologies to exploit a feature combination of a machine learning for hypothesis testing. Then, the existing type-labeled instances are used for classifying type-unlabeled new instances with the validation of testing dataset errors. Finally, these newly type-labeled instances are further forwarded to the structured ontologies to empower the ontology and rule learning. The proposed concepts have been tested and verified for intelligent security with the real KDD CUP 1999 datasets.