Graph-Based Learning: Method and Application

A longstanding open problem in machine learning and data science is deter-
mining the quality of data for training a learning algorithm, e.g., a classifier.
Several algorithms to learn the intrinsic quality of data directly from a data
sample have been proposed in the past. In addition to accuracy, computational
and sample complexity of learning algorithms is important as it affects the practicality of data quality predictors. This talk will present a geometric analytical
framework, inspired by geometry of graphs and information theoretic measures,
for estimating empirical data quality measures from data. The framework will
be illustrated for applications in multi-class classification and data mining.
Salimeh is a postdoctoral research fellow in EECS department at the Univer-
sity of Michigan, working with Prof. Alfred O. Hero. Prior to joining Michigan,
she was CAPES-PNPD funder postdoctoral fellow at Federal Univesrity of Sao
Carlos (UFSCar), Brazil, in 2014 and 2015. She was visiting scholar at Polytech-
nic University of Turin, Italy between 2011 and 2013. She received her Ph.D.
in Inferential Statistics from the Ferdowsi University of Mashhad (FUM), Iran
in 2013, her M.S. degree in Mathematical Statistics (2007) and B.S. degree in
Statistics (2004), both again from FUM. She works on the area of Machine
Learning, Statistical Inference, and Data Science. Her research focuses on de-
veloping and analyzing methods in graph-based learning and high-dimensional
and massive data inference problems.