Robust Optimization and Applications in Machine Learning
Part 2: Robust Classification
Data matrix
Classification problems
What is a linear classifier?
Separable data
Non-separable data
Loss functions
Two specific loss functions
Generalization error and regularization
Regularization and Sparsity
Robust classification
Formulation of robustness approach
Non-separable case
Link with worst-case loss minimization
Box uncertainty model
Formulation
Link with worst-case loss minimization
Our findings so far
Part 2: Robust Classification
Classification with interval data
Robust classification: main idea
Main results
Part 2: Robust Classification
Robust classification with hinge loss
Bound on robust SVM
Part 2: Robust Classification
Robust LR classification
Robust LR: dual
Moment matching
Part 2: Robust Classification
Minimax probability machine
Problem statement
Problem formulation
Marhsall and Olkin’s result ? ?
SOCP formulation
Dual problem
Geometric interpretation
Solving the problem
Robustness to estimation errors
Robust MPM
Formulation of Robust MPM Lemma
R-MPM: A Specific Uncertainty Model (1)
R-MPM: A Specific Uncertainty Model (2)
Robust MPM: Estimation Errors in Means
Rost MPM: Estimation Errors in Covariance
R-MPM: putting everything together
Part 2: summary