10 likes | 173 Views
Fast Algorithms on Imperfect Heterogeneous Distributed Data for Interactive Analysis. Large-scale Nonnegative Matrix Factorization For better interpretability & quality. Hawke’s Process Predict future events. Capability Topic modeling Clustering Dimension reduction Outlier detection
E N D
Fast Algorithms on Imperfect Heterogeneous Distributed Data for Interactive Analysis Large-scale Nonnegative Matrix Factorization For better interpretability & quality Hawke’s Process Predict future events Capability Topic modeling Clustering Dimension reduction Outlier detection Recommendation Spatio-temporal modeling Data ≅ x GISR: Topic /Network Discovery Kiva: Loan Recommendation CIDR: Major Pattern & Outlier Topical + Loan metadata + Default/Delinq. + Temporal + Team info * Seoul has a unique pattern for gaming. Atlanta has major credit card transaction traffic. * Teams influence only active lenders while non-paid loans impact only inactive lenders. * Revealed key terrorists and bomb-related activities from communications data.