|
Activity Number:
|
288
- SLDS CSpeed 5
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #317979
|
|
|
Title:
|
A Penalized Model-Based Coclustering Algorithm
|
|
Author(s):
|
Chenchen Ma* and Gongjun Xu and Ji Zhu
|
|
Companies:
|
University of Michigan and University of Michigan and University of Michigan
|
|
Keywords:
|
Co-clustering;
Latent block model;
Model selection;
EM algorithm
|
|
Abstract:
|
Co-clustering is a data mining technique which allows simultaneous clustering of the rows and columns of a matrix. In this project, based on the latent block model, we propose a two-stage penalized likelihood approach with two additional penalty terms: one is log-type penalty on the proportion parameters and the other is the truncated Lasso penalty on the differences of block parameters. We develop an efficient EM algorithm using DC programming and ADMM method. The proposed algorithm is capable of automatically selecting numbers of clusters and learning latent checkerboard structures. Clustering consistency is also guaranteed under mild conditions. We demonstrate the good performance of the proposed method through comprehensive simulation studies and a real data set from a personality factors test.
|
Authors who are presenting talks have a * after their name.
