Abstract: This paper proposes a novel secular knowledge representation and learning framework to proposed large-scale secular signature mining of longitudinal heterogeneous occasional data. The framework allows the presentation, extra4ction, and mining of high order latent occasion event structure and relationships between single and many sequences. The prescribed data representation maps the heterogeneous sequences to a image by encoding occasions as a structured spatial-secular shape process. We have suggested clinical assessment for naked interactive knowledge discovery in large electronic health record databases.
Keywords: Secular signature mining, sparse coding, dictionary positive matrix factorization.