Distributed Coordinate Descent for L1-regularized Logistic Regression
By Ilya Trofimov, A. Genkin. 2014
15 citations recorded by the source corpus.
Solving logistic regression with L1-regularization in distributed settings is an important problem. This problem arises when training dataset is very large and cannot fit the memory of a single machine. We present d-GLMNET, a new algorithm solving logistic regression with L1-regularization in the distributed settings. We empirically show that it is superior over distributed online learning via truncated gradient.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Martin Jaggi
- Virginia Smith
- Peter Richtárik
- Jakub Konecný
- Martin Takác
- Chenxin Ma
- Michael I. Jordan
- Sebastian U. Stich
In Computer Vision and Pattern Recognition
- Virginia Smith
- Martin Takác
- Martin Jaggi
- Michael I. Jordan
- Chenxin Ma
- Shan-Wei Lin
- Dr. Matthias Seeger
- Shou-De Lin
Citations in the corpus
- CoCoA: A General Framework for Communication-Efficient Distributed Optimization (2016, background): “While Mahajan et al. (2014) allows a fixed accuracy for these subproblems, but not arbitrary approximation quality Θ as in our framework, the works of Trofimov and Genkin (2014); Yuan et al. (2012); and Yen et al. (2015) assume that the quadratic subproblems are solved exactly.”
- A Dual Augmented Block Minimization Framework for Learning with Limited Memory (2015, methodology): “) [7, 10] that only apply to ERM with L2-regularizer or some other distributed method designed for some specific loss function [19].”
- Stochastic Dual Coordinate Ascent with Adaptive Probabilities (2015, unknown): “O C ] 27 F”
- Distributed Optimization for Non-Strongly Convex Regularizers (2016, methodology): “On the theory side, its analysis [15, 17] is not an explicit convergence rate but only asymptotic, as the quadratic upper bounds are not explicitly controlled for safety as with our σ′.”
- L1-Regularized Distributed Optimization: A Communication-Efficient Primal-Dual Framework (2015, background): “While [18] allows a fixed accuracy for these subproblems—but not arbitrary approximation quality Θ as in our framework—the work of [29, 34, 31] assumes that the quadratic subproblems are solved exactly.”