Abstract:
Teasing out interesting relationships buried within volumes of data is one of the most fundamental challenges in data science research. Increasingly, researchers and practitioners are interested in understanding how individual pieces of information are organized and interact in order to discover the fundamental principles that underlie complex physical or social phenomena. Arguably, the most prescient task in the study of such systems is the identification, extraction, and representation of the small substructures that, in aggregate, describe the underlying phenomenon encoded by the graph. Furthermore, this opens the door for researchers to design tools for tasks like anomaly or fraud detection as well as anonymizing social network data. The research described in this talk makes significant progress in filling the void in current research to find scalable, and interpretable graph modeling methods by leveraging the new-found link between formal language theory, graph theory, and data mining. We achieve this by adopting the formalism of vertex replacement grammars, which lets us describe the
LEGO-like building blocks of a graph. This helps us bridge the gap between subgraph mining and graph generation to create a new suite of models and tools that can not only create informative models of real-world data, but also generate, extrapolate, and infer new graphs in a precise, principled way.
Bio:
Satyaki Sikdar is a Ph.D. Candidate in the Department of Computer Science & Engineering at the University of Notre Dame advised by Dr. Tim Weninger. His research is at the intersection of graph theory, formal language theory, and data mining. His work has been published in IEEE ICDM, IEEE BigData, WSDM, and KAIS. He is presently the instructor of record for CSE 20110: Discrete Mathematics, a required course for Computer Science majors.
Faculty Host: Brad Solomon
