Missing Value Imputation

Auto-BI helps end-users by accurately predicting BI models given a set of input tables by developing a principled graph-based optimization problem in Auto-BI that considers both local join prediction and global schema-graph structures. Extensive experiments on 1000 real test cases suggest that Auto-BI is both efficient and accurate, achieving over 90% F1-score when evaluated against ground-truth BI models that humans design.

ZIP develops a query-time missing value imputation framework that minimizes the joint costs of imputation and query execution. QUIP outperforms the state-of-the-art ImputeDB by 2 to 10 times on different query sets and data sets, and achieves the order-of-magnitudes improvement over offline approach.