Retrospective Extraction of Visual and Logical Insights for Ontology-based interpretation of Neural Networks