DocumentCode
114241
Title
Dynamic stochastic optimization
Author
Wilson, Craig ; Veeravalli, Venugopal ; Nedic, Angelia
Author_Institution
Univ. of Illinois at Urbana-Champaign, Urbana, IL, USA
fYear
2014
fDate
15-17 Dec. 2014
Firstpage
173
Lastpage
178
Abstract
A framework for sequentially solving stochastic optimization problems with stochastic gradient descent is introduced. Two tracking criteria are considered, one based on being accurate with respect to the mean trajectory and the other based on being accurate in high probability (IHP). An off-line optimization problem is solved to find the constant step size and number of iterations to achieve the desired tracking accuracy. Simulations are used to confirm that this approach provides the desired tracking accuracy.
Keywords
gradient methods; optimisation; probability; stochastic processes; tracking; IHP; dynamic stochastic optimization; in high probability; mean trajectory; off-line optimization problem; stochastic gradient descent; tracking criteria; Accuracy; Algorithm design and analysis; Analytical models; Least squares approximations; Markov processes; Optimization; adaptive optimization; gradient methods; stochastic optimization;
fLanguage
English
Publisher
ieee
Conference_Titel
Decision and Control (CDC), 2014 IEEE 53rd Annual Conference on
Conference_Location
Los Angeles, CA
Print_ISBN
978-1-4799-7746-8
Type
conf
DOI
10.1109/CDC.2014.7039377
Filename
7039377
Link To Document