Incremental Computation of Information Gain in Temporal Relational Streams.

This paper studies a highly efficient statistical method in identifying anomalies and outliers in relational data streams. We compute information gain as a statistical measure of relational values in order to determine their significance. In contrast to more sophisticated model based methods, our algorithm can maintain the information gain measures incrementally in $\mathcal{O} (1)$ cost, thus allowing an online monitoring of temporal relational streams.