Activity Number:

253
 Contributed Poster Presentations: Section on Statistical Computing

Type:

Contributed

Date/Time:

Monday, July 30, 2018 : 2:00 PM to 3:50 PM

Sponsor:

Section on Statistical Computing

Abstract #329643


Title:

Statistical Techniques to Improve Random Projections and Other Similar Algorithms

Author(s):

Keegan Kang* and Weipin Wong and Haikal Yeo

Companies:

Singapore University Of Technology And Design and Singapore University Of Technology and Design and Independent

Keywords:

dimension reduction;
random projections;
bayesian prior;
control variates;
maximum likelihood estimates

Abstract:

Random projections are used to approximately preserve distances in high dimensional space. A transformation involving random variables is applied to a dataset, mapping it to a smaller subspace. The original distance is then estimated via an unbiased estimator. We show how we can apply statistical techniques such as maximum likelihood estimates, control variates, and Bayesian priors to improve the accuracy of such estimates with negligible computational cost. We also show how our techniques can be generalized to similar algorithms as well.
