BriefGPT.xyz
May, 2011
L1正则化损失函数的并行坐标下降优化算法
Parallel Coordinate Descent for L1-Regularized Loss Minimization
HTML
PDF
Joseph K. Bradley, Aapo Kyrola, Danny Bickson, Carlos Guestrin
TL;DR
采取Shotgun算法并在Lasso回归和稀疏逻辑回归方面进行了综合的理论和实证研究，证明了该算法在大规模问题上的扩展性表现非常优越。
Abstract
We propose Shotgun, a parallel
coordinate descent
algorithm for minimizing
l1-regularized losses
. Though
coordinate descent
seems inherent
→