Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


thesis.pdf (564.38 KB)

ETD Abstract Container

Abstract Header

Sparse Multinomial Logistic Regression via Approximate Message Passing

Byrne, Evan Michael
http://rave.ohiolink.edu/etdc/view?acc_num=osu1437416281

Abstract Details

2015, Master of Science, Ohio State University, Electrical and Computer Engineering.
For the problem of multi-class linear classification and feature selection, we propose new approximate message passing algorithms based on Hybrid Generalized Approximate Message Passing (Hybrid-GAMP) to train a multinomial logistic regression model. We propose both maximum a posteriori (MAP) and approximate minimum mean-square error (MMSE) estimators of the weight vectors. Then we design simplified variants of these algorithms that lead to significantly faster runtimes and improved numerical robustness. Our algorithms are able to handle the case where the number of features far exceeds the number of training examples through the use of sparsity-promoting prior distributions on the weight vectors. Additionally, our algorithms are able to take advantage of existing expectation-maximization (EM) and Stein's Unbiased Risk Estimate (SURE) methods to tune their parameters online. Finally, we demonstrate our algorithms' performance on both synthetic and real datasets.
Philip Schniter (Advisor)
Lee Potter (Committee Member)
84 p.
Electrical Engineering
multinomial logistic regression; multiclass linear classification; approximate message passing; feature selection

Recommended Citations

Citations

  • Byrne, E. M. (2015). Sparse Multinomial Logistic Regression via Approximate Message Passing [Master's thesis, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1437416281

    APA Style (7th edition)

  • Byrne, Evan. Sparse Multinomial Logistic Regression via Approximate Message Passing. 2015. Ohio State University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1437416281.

    MLA Style (8th edition)

  • Byrne, Evan. "Sparse Multinomial Logistic Regression via Approximate Message Passing." Master's thesis, Ohio State University, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=osu1437416281

    Chicago Manual of Style (17th edition)

osu1437416281
489
© 2015, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.