Title: STOCHASTIC APPROXIMATIVE LASSO FOR MASSIVE DATA VIA SMALL SAMPLE BAGGING

Abstract: This talk will present a recent work on what to expect on
small sample bagging in high dimensional spaces when traditional
estimators are not achievable.  When dealing with Massive Data with n
of the order of millions and p of the order of thousands, computing
Lasso estimator for example becomes computationally intractable task.
We propose a procedure based on bagging small sample Lasso estimators
and variance stabilization technique to approximate Lasso
solution. Without requiring additional conditions we show that the
approximation is good in predictive or support recovery sense but
almost never simultaneously. The algorithm is used to analyze highly
complex and rather new Red Cross data, with 5M entries and almost
impossible task of analyzing features for successful donor retention
strategies from marketing perspective.


Bio: 
Jelena Bradic is an Assistant Professor at the Department of
Mathematics, UCSD. She received her Ph.D. in Operations Research and
Financial Engineering from Princeton in Spring 2011 with a
specialization in Statistics and Applied Probability under the
direction of Jianqing Fan. Her research is in high dimensional
statistics, stochastic optimization, asymptotic theory, robust
statistics, functional genomics and biostatistics.  She has received
LAHA award from Institute of Mathematical Statistics in 2010.