Training Experimentally Robust and Interpretable Binarized Regression Models Using Mixed-Integer Programming

被引:0
作者
Tule, Sanjana [1 ]
Le, Nhi Ha Lan [1 ]
Say, Buser [1 ]
机构
[1] Monash Univ, Melbourne, Vic, Australia
来源
2022 IEEE SYMPOSIUM SERIES ON COMPUTATIONAL INTELLIGENCE (SSCI) | 2022年
关键词
robust machine learning; interpretable machine learning; mixed-integer programming;
D O I
10.1109/SSCI51031.2022.10022152
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In this paper, we explore model-based approach to training robust and interpretable binarized regression models for multiclass classification tasks using Mixed-Integer Programming (MIP). Our MIP model balances the optimization of prediction margin and model size by using a weighted objective that: minimizes the total margin of incorrectly classified training instances, maximizes the total margin of correctly classified training instances, and maximizes the overall model regularization. We conduct two sets of experiments to test the classification accuracy of our MIP model over standard and corrupted versions of multiple classification datasets, respectively. In the first set of experiments, we show that our MIP model outperforms an equivalent Pseudo-Boolean Optimization (PBO) model and achieves competitive results to Logistic Regression (LR) and Gradient Descent (GD) in terms of classification accuracy over the standard datasets. In the second set of experiments, we show that our MIP model outperforms the other models (i.e., GD and LR) in terms of classification accuracy over majority of the corrupted datasets. Finally, we visually demonstrate the interpretability of our MIP model in terms of its learned parameters over the MNIST dataset. Overall, we show the effectiveness of training robust and interpretable binarized regression models using MIP.
引用
收藏
页码:838 / 845
页数:8
相关论文
共 26 条
[1]   CONSISTENT MANIFOLD REPRESENTATION FOR TOPOLOGICAL DATA ANALYSIS [J].
Berry, Tyrus ;
Sauer, Timothy .
FOUNDATIONS OF DATA SCIENCE, 2019, 1 (01) :1-38
[2]   SmcHD1, containing a structural-maintenance-of-chromosomes hinge domain, has a critical role in X inactivation [J].
Blewitt, Marnie E. ;
Gendrel, Anne-Valerie ;
Pang, Zhenyi ;
Sparrow, Duncan B. ;
Whitelaw, Nadia ;
Craig, Jeffrey M. ;
Apedaile, Anwyn ;
Hilton, Douglas J. ;
Dunwoodie, Sally L. ;
Brockdorff, Neil ;
Kay, Graham F. ;
Whitelaw, Emma .
NATURE GENETICS, 2008, 40 (05) :663-669
[3]  
Braun D, 2017, 18TH ANNUAL MEETING OF THE SPECIAL INTEREST GROUP ON DISCOURSE AND DIALOGUE (SIGDIAL 2017), P174
[4]  
Bühlmann P, 2011, SPRINGER SER STAT, P1, DOI 10.1007/978-3-642-20192-9
[5]  
Clark P., 1989, Machine Learning, V3, P261, DOI 10.1007/BF00116835
[6]  
Collobert R, 2011, J MACH LEARN RES, V12, P2493
[7]  
Romano JD, 2021, Arxiv, DOI arXiv:2012.00058
[8]  
Deng L, 2013, INT CONF ACOUST SPEE, P8599, DOI 10.1109/ICASSP.2013.6639344
[9]  
Elffers J, 2018, PROCEEDINGS OF THE TWENTY-SEVENTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, P1291
[10]  
Geiger L., 2020, Journal of Open Source Software, V5, P1746, DOI [DOI 10.21105/JOSS.01746, 10.21105/joss.01746]