An Interior-Point Method for Large-Scale l1-Regularized Logistic Regression

By Kwangmoo Koh, Seung-Jean Kim, Stephen P. Boyd. Journal of machine learning research, 2007

749 citations recorded by the source corpus.

Subfields: Computational Theory and Mathematics, Mathematical Physics

View Paper: https://www.semanticscholar.org/paper/e7f716ab4e23cfb423a4f6b19363678d922a76f3

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Mathematical Physics

Citations in the corpus

Parallel Coordinate Descent for L1-Regularized Loss Minimization (2011, result): “On L1 logreg (Koh et al., 2007) and CDN (Yuan et al., 2010), our results qualitatively matched their survey.”
Conditional random fields for activity recognition (2007, methodology): “Combined with the warm start technique used by [50], this reparameterization allows us to train CRFs with an `1 penalty [109].”
Large-scale sparse logistic regression (2009, methodology): “In this experiment, we compare the proposed Lassplore algorithm with two recent solvers: ProjectionL1 [30] and l1-logreg [16] in terms of the computational time for solving the sparse logistic regression.”
An empirical study of context in object detection (2009, methodology): “We use logistic regression [19] to train two separate object presence classifiers based on each descriptor.”
High-Dimensional Graphical Model Selection Using $\ell_1$-Regularized Logistic Regression (2008, methodology): “In all cases, we solved the ℓ 1 -regularized logistic regression using special purpose interior-point code developed and described by Koh et al. (2007).”
Face Liveness Detection from a Single Image with Sparse Low Rank Bilinear Discriminative Model (2010, background): “The sparse logistic regression [9,14] add a `1-norm regularization to the loss to avoid overfitting, i.”
Reinforcement Learning Transfer Using a Sparse Coded Inter-task Mapping (2011, methodology): “This optimization problem guarantees that the attained activations are as sparse as possible and is solved using the interior point methods [4].”
Recent Advances of Large-Scale Linear Classification (2012, methodology): “• L1-regularized LR: Most methods solve the primal form, for example, an interior-point method (l1 logreg [37]), (block) coordinate descent methods (BBR [38] and CGD [39]), a quasi-Newton method (OWL-QN [40]), Newton-type methods (GLMNET [41] and LIBLINEAR [22]), and a Nesterov’s method (SLEP [42]).”
Blitz: A Principled Meta-Algorithm for Scaling Sparse Optimization (2015, methodology): “2 (Koh et al., 2007): Interior point method for sparse logistic regression written in C.”
A Survey of Sparse Representation: Algorithms and Applications (2015, methodology): “We analyze and compare the performance of sparse representation with the most typical algorithms: OMP [37], l1_ls [76], PALM [89], FISTA [82], DALM [89], homotopy [99] and TPTSR [9].”
Learning locomotion over rough terrain using terrain templates (2009, methodology): “We run an efficient implementation of l1-regularized logistic regression (LR) [11] on the pairwise difference feature vector data (xc − xi).”
FastXML: a fast, accurate and stable tree-classifier for extreme multi-label learning (2014, background): “This problem has been extensively studied in the literature [3,24,37].”
Heterogeneous multitask learning with joint sparsity constraints (2009, positive): “For example, an interior point method and a preconditioned conjugate gradient algorithm have been used to solve a large-scale L1-regularized linear regression and logistic regression [8].”
Exponentiated Gradient Algorithms for Conditional Random Fields and Max-Margin Markov Networks (2008, methodology): “A similar initialization method is used in Koh et al. (2007) in the context of `1 regularized logistic regression.”
Incrementally Updated Gradient Methods for Constrained and Regularized Optimization (2013, background): “objective value, and the CPU time (in seconds) for solving randomly generated problems described in [32] to see the effect of the choice of the number (i.”
Towards Transparent Systems: Semantic Characterization of Failure Modes (2014, methodology): “We use interior based method for this optimization [38].”