With the advent of big data, there is a growing need to provide a structured and organized view of the data for effective search, browsing and data mining. This talk focuses on how to exploit human provided structures in the learning process; specifically, how to leverage hierarchical or graph based dependencies between class-labels, in a tractable manner, to improve the classification performance. I will present a suite of algorithms and optimization methods that can seamlessly learn billions of parameters across hundreds of thousands of class-labels in a matter of several hours.
Siddharth Gopal is a PhD student at the Language Technologies Institute at Carnegie Mellon University. His research interests has centered on information retrieval and machine learning. He is currently interested in developing scalable methods for learning in the presence of large number of classes, inferring structures from unlabeled data and recovering partial and incomplete structures based on user preferences.
Faculty Host: Yiming Yang