Outlier or anomaly detection in large graph data sets is a fundamental task in data science, with broad applications. In this talk, I will present three novel frameworks we have developed with our collaborators for graph anomaly analysis: (i) Graph OLAP. Graph OLAP aims to provide a systematic framework to perform composite structure and information analysis in networks. (ii) Graph Anomaly. Under this term, we have developed a few interesting concepts such as graph iceberg, structural correlation, graph association and aggregation. (iii) Query-based Outlier Detection. As users have their own intuition about which kind of outliers they are interested in, a system shall allow users to specify outlier queries flexibly. I will introduce the concept of query-based outlier in heterogeneous information networks and a declarative query language. In the end of my talk, I will discuss a few existing problems on graph anomaly detection.

This talk will present RAPT, a new predictive modeling framework for identifying rare classes in complete absence of labeled data. The RAPT framework is designed to use imperfectly annotated training data to learn classification models in the absence of expert-annotated training samples. Our results show that, under some reasonable assumptions, the classifiers trained from imperfectly labeled training data using the RAPT approach have performance comparable to the classification models trained using expert-annotated training data. This capability of learning from imperfect supervision is advantageous in a wide range of applications where the target class of interest is relatively rare and obtaining a precise labeling of even a small number of training samples is infeasible. The talk will present the application of the RAPT framework for creating historical maps of forest fires from satellite data for the tropical forests. This new forest fire product identifies approximately 1 million sq. km. of burned areas in the tropical forests in South America and South-east Asia during years 2001-2014, which is more than double of the total burned area reported by the state-of-art NASA products. We show validation of these results using burn-scars visible in satellite images, including high resolution Landsat images, to confirm the veracity of the previously unreported forest fires.

Here is the list of discussion questions.