Stochastic Gradient Descent on Riemannian Manifolds

被引:366
作者
Bonnabel, Silvere [1 ]
机构
[1] Mines ParisTech, Robot Lab, Math & Syst, F-75272 Paris, France
关键词
Nonlinear identification; Riemannian geometry; stochastic approximation; OPTIMIZATION;
D O I
10.1109/TAC.2013.2254619
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Stochastic gradient descent is a simple approach to find the local minima of a cost function whose evaluations are corrupted by noise. In this paper, we develop a procedure extending stochastic gradient descent algorithms to the case where the function is defined on a Riemannian manifold. We prove that, as in the Euclidian case, the gradient descent algorithm converges to a critical point of the cost function. The algorithm has numerous potential applications, and is illustrated here by four examples. In particular a novel gossip algorithm on the set of covariance matrices is derived and tested numerically.
引用
收藏
页码:2217 / 2229
页数:13
相关论文
共 37 条
[1]  
Absil PA, 2008, OPTIMIZATION ALGORITHMS ON MATRIX MANIFOLDS, P1
[2]   RIEMANNIAN Lp CENTER OF MASS: EXISTENCE, UNIQUENESS, AND CONVEXITY [J].
Afsari, Bijan .
PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY, 2011, 139 (02) :655-673
[3]  
Amari S. I., 1998, NATURAL GRADIENT WOR
[4]  
[Anonymous], 1998, ONLINE ALGORITHMS ST
[5]  
[Anonymous], 1983, SUBSPACE METHODS PAT
[6]  
[Anonymous], 2008, P 2008 IEEE RAD C RO
[7]   Stochastic algorithms for computing means of probability measures [J].
Arnaudon, Marc ;
Dombry, Clement ;
Phan, Anthony ;
Yang, Le .
STOCHASTIC PROCESSES AND THEIR APPLICATIONS, 2012, 122 (04) :1437-1455
[8]   ANALYSIS OF STOCHASTIC-APPROXIMATION SCHEMES WITH DISCONTINUOUS AND DEPENDENT FORCING TERMS WITH APPLICATIONS TO DATA COMMUNICATION ALGORITHMS [J].
BENVENISTE, A ;
GOURSAT, M ;
RUGET, G .
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 1980, 25 (06) :1042-1058
[9]  
Bonnabel S., 2010, P INT S MATH THEOR N
[10]  
Bonnabel S., 2011, P 13 C GRETSI BORD