Research Buffet
Wednesday, October 26, 2011
3147 MEB
Refreshments 3:40 p.m.
Lecture 4:00 p.m.
Alex Clemmer
Title: Finding Evil Executives in the Enron Email Corpus: Methods and Empirical Problems in Vertex Nomination
Wednesday, October 26, 2011
3147 MEB
Refreshments 3:40 p.m.
Lecture 4:00 p.m.
Alex Clemmer
Title: Finding Evil Executives in the Enron Email Corpus: Methods and Empirical Problems in Vertex Nomination
Abstract
Probably the most common existential case of graph problems found in the real world is the case where we have strong verified knowledge of a useful set of attributes for only some of the vertices. In such cases, it is frequently true that we care particularly about one feature, or one set of features. The process of using information derived from the set of vertices for which we have such strong verified knowledge, and then nominating vertices that are likely to have this attribute, is called Vertex nomination [Coppersmith and Priebe 2011].
This set of problems has a wide range of applications, but in this talk I will speak mainly about it in terms of the problem of finding "evil" executives in the Enron email corpus. Specifically, I will give an overview of our approach to nomination in this context, as well as a brief note on the empirical problems inherent to the problem of vertex nomination in general.
Chad Brubaker
Title: Adaptive Sampling for Large-data Multidimensional Scaling
Probably the most common existential case of graph problems found in the real world is the case where we have strong verified knowledge of a useful set of attributes for only some of the vertices. In such cases, it is frequently true that we care particularly about one feature, or one set of features. The process of using information derived from the set of vertices for which we have such strong verified knowledge, and then nominating vertices that are likely to have this attribute, is called Vertex nomination [Coppersmith and Priebe 2011].
This set of problems has a wide range of applications, but in this talk I will speak mainly about it in terms of the problem of finding "evil" executives in the Enron email corpus. Specifically, I will give an overview of our approach to nomination in this context, as well as a brief note on the empirical problems inherent to the problem of vertex nomination in general.
Chad Brubaker
Title: Adaptive Sampling for Large-data Multidimensional Scaling
Abstract
My talk will focus on using random sampling of points to run a streaming variant of MDS that works for large datasets. A quick introduction to MDS and its applications will be provided but for the most part I will talk about my work on streaming and random sampling with a small note on multiprocessing work.
My talk will focus on using random sampling of points to run a streaming variant of MDS that works for large datasets. A quick introduction to MDS and its applications will be provided but for the most part I will talk about my work on streaming and random sampling with a small note on multiprocessing work.