English

Learning Sparse Classifiers: Continuous and Mixed Integer Optimization Perspectives

Machine Learning 2021-06-08 v2 Machine Learning Optimization and Control Computation

Abstract

We consider a discrete optimization formulation for learning sparse classifiers, where the outcome depends upon a linear combination of a small subset of features. Recent work has shown that mixed integer programming (MIP) can be used to solve (to optimality) 0\ell_0-regularized regression problems at scales much larger than what was conventionally considered possible. Despite their usefulness, MIP-based global optimization approaches are significantly slower compared to the relatively mature algorithms for 1\ell_1-regularization and heuristics for nonconvex regularized problems. We aim to bridge this gap in computation times by developing new MIP-based algorithms for 0\ell_0-regularized classification. We propose two classes of scalable algorithms: an exact algorithm that can handle p50,000p\approx 50,000 features in a few minutes, and approximate algorithms that can address instances with p106p\approx 10^6 in times comparable to the fast 1\ell_1-based algorithms. Our exact algorithm is based on the novel idea of \textsl{integrality generation}, which solves the original problem (with pp binary variables) via a sequence of mixed integer programs that involve a small number of binary variables. Our approximate algorithms are based on coordinate descent and local combinatorial search. In addition, we present new estimation error bounds for a class of 0\ell_0-regularized estimators. Experiments on real and synthetic data demonstrate that our approach leads to models with considerably improved statistical performance (especially, variable selection) when compared to competing methods.

Keywords

Cite

@article{arxiv.2001.06471,
  title  = {Learning Sparse Classifiers: Continuous and Mixed Integer Optimization Perspectives},
  author = {Antoine Dedieu and Hussein Hazimeh and Rahul Mazumder},
  journal= {arXiv preprint arXiv:2001.06471},
  year   = {2021}
}

Comments

To appear in JMLR