Brent Castle
PhD student
School of Informatics and Computing
Indiana University
Interests
Stochastic Optimization. I am particularly interested in optimizing objective functions that must be estimated by simulations. One common scenario is estimating the parameters of a stochastic process. My advisor (Michael Trosset) and I are developing quasi-Newton methods to address such problems, however, the methodology can be applied to more general stochastic optimization problems.
Machine Learning. I am interested in methods for combining information from disparate sources with the purpose of classification. To form an intermediate representation of the data I have adopted several embedding strategies for combining multiple sets of dissimilarities.
A Nonmetric Embedding Approach to Testing for Matched Pairs. 2011 IU Department of Statistics TR. (with Michael W. Trosset and Carey E. Priebe) [pdf]
Fast Euclidean Embedding of Ordinal Nearest Neighbor Graphs. 2010 Joint Statistical Meetings. (with Michael W. Trosset) [pdf]
Feature Extraction for Multiple Kernel Learning. 2009 IU Department of Statistics TR. (with Minh Tang and Michael W. Trosset) [pdf]
Posters
Fast Euclidean Embedding of Ordinal Nearest Neighbor Graphs. 2010 Joint Statistical Meetings. (with Michael W. Trosset) [pdf]
Learning from Heterogeneous Data Sources by Combining Dissimilarities. 2010 Conference on Nonparametric Statistics and Statistical Learning. (with Michael W. Trosset)
Combining Disparate Information by Nonmetric Multidimensional Scaling. 2008 IU CSGSA Poster Session (with Michael W. Trosset) [pdf]
Talks
Quasi-Newton Methods for Stochastic Optimization With Application to Simulation-Based Parameter Estimation. 2011 MOPTA. (with Michael W. Trosset) [pdf]
Learning from Heterogeneous Data Sources by Combining Dissimilarities. 2010 Interface. (with Michael W. Trosset) [pdf]
Combining Disparate Information by Nonmetric Multidimensional Scaling. 2008 Interface. (with Michael W. Trosset) [pdf]
