Improved Hessian estimation for adaptive random directions stochastic approximation

Reddy, D Sai Koti and Prashanth, L A and Bhatnagar, Shalabh (2017) Improved Hessian estimation for adaptive random directions stochastic approximation. In: 55th IEEE Conference on Decision and Control (CDC), DEC 12-14, 2016, Las Vegas, NV, pp. 3682-3687.

PDF 2016_Iee_55th_con_Dec_Con_3682.pdf - Published Version Restricted to Registered users only Download (363kB) | Request a copy

Abstract

We propose an improved Hessian estimation scheme for the second-order random directions stochastic approximation (2RDSA) algorithm I]. The proposed scheme, inspired by 121, reduces the error in the Hessian estimate by (i) incorporating a zero-mean feedback term; and (ii) optimizing the step-sizes used in the Hessian recursion of 2RDSA. We prove that 2RDSA with our Hessian improvement scheme (2RDSAIII) converges asymptotically to the true Hessian. The advantage with 2RDSA-III is that it requires only 75% of the simulation cost per-iteration for 2SPSA with improved Hessian estimation (2SPSA-IH) 2]. Numerical experiments show that 2RDSA-IH outperforms both 2SPSA-Ill and 2RDSA without the improved hessian estimation scheme.

Item Type:	Conference Proceedings
Series.:	IEEE Conference on Decision and Control
Publisher:	IEEE, 345 E 47TH ST, NEW YORK, NY 10017 USA
Additional Information:	55th IEEE Conference on Decision and Control (CDC), Las Vegas, NV, DEC 12-14, 2016
Department/Centre:	Division of Electrical Sciences > Computer Science & Automation
Date Deposited:	10 Jun 2017 04:41
Last Modified:	10 Jun 2017 04:41
URI:	http://eprints.iisc.ac.in/id/eprint/57204

Actions (login required)

View Item