@Article{2009passch,
	author      =   {R. Pasupathy and B. W. Schmeiser},
	title       =   {Retrospective-approximation algorithms for multidimensional stochastic root-finding problems},
	journal     =   {ACM TOMACS},
	year        =   {2009},
	volume      =   {19},
	number      =   {2},
	pages       =   {5:1--5:36},
	doi = {10.1145/1502787.1502788},
	url = {http://web.ics.purdue.edu/~pasupath/PAPERS/2009passch.pdf},
	keywords = {retrospective approximation, stochastic root-finding},
	abstract = {The stochastic root-finding problem (SRFP) is that of solving a nonlinear system of equations using only a simulation that provides estimates of the functions at requested points. Equivalently, SRFPs seek locations where an unknown vector function attains a given target using only a simulation capable of providing estimates of the function. SRFPs find application in a wide variety of physical settings. We develop a family of retrospective-approximation (RA) algorithms called Bounding RA that efficiently solves a certain class of multidimensional SRFPs. During each iteration, Bounding RA generates and solves a sample-path problem by identifying a polytope of stipulated diameter, with an image that bounds the given target to within stipulated tolerance. Across iterations, the stipulations become increasingly stringent, resulting in a sequence of shrinking polytopes that approach the correct solution. Efficiency results from: (i) the RA structure, (ii) the idea of using bounding polytopes to exploit problem structure, and (iii) careful step-size and direction choice during algorithm evolution. Bounding RA has good finite-time performance that is robust with respect to the location of the initial solution, and algorithm parameter values. Empirical tests suggest that Bounding RA outperforms Simultaneous Perturbation Stochastic Approximation (SPSA), which is arguably the best-known algorithm for solving SRFPs.}}




%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% PRELIMINARY BOOK CITATIONS

{"_id":"3bxWRof2s5SYG6Wcp","bibbaseid":"pasupathy-schmeiser-retrospectiveapproximationalgorithmsformultidimensionalstochasticrootfindingproblems-2009","author_short":["Pasupathy, R.","Schmeiser, B. W."],"bibdata":{"bibtype":"article","type":"article","author":[{"firstnames":["R."],"propositions":[],"lastnames":["Pasupathy"],"suffixes":[]},{"firstnames":["B.","W."],"propositions":[],"lastnames":["Schmeiser"],"suffixes":[]}],"title":"Retrospective-approximation algorithms for multidimensional stochastic root-finding problems","journal":"ACM TOMACS","year":"2009","volume":"19","number":"2","pages":"5:1–5:36","doi":"10.1145/1502787.1502788","url":"http://web.ics.purdue.edu/~pasupath/PAPERS/2009passch.pdf","keywords":"retrospective approximation, stochastic root-finding","abstract":"The stochastic root-finding problem (SRFP) is that of solving a nonlinear system of equations using only a simulation that provides estimates of the functions at requested points. Equivalently, SRFPs seek locations where an unknown vector function attains a given target using only a simulation capable of providing estimates of the function. SRFPs find application in a wide variety of physical settings. We develop a family of retrospective-approximation (RA) algorithms called Bounding RA that efficiently solves a certain class of multidimensional SRFPs. During each iteration, Bounding RA generates and solves a sample-path problem by identifying a polytope of stipulated diameter, with an image that bounds the given target to within stipulated tolerance. Across iterations, the stipulations become increasingly stringent, resulting in a sequence of shrinking polytopes that approach the correct solution. Efficiency results from: (i) the RA structure, (ii) the idea of using bounding polytopes to exploit problem structure, and (iii) careful step-size and direction choice during algorithm evolution. Bounding RA has good finite-time performance that is robust with respect to the location of the initial solution, and algorithm parameter values. Empirical tests suggest that Bounding RA outperforms Simultaneous Perturbation Stochastic Approximation (SPSA), which is arguably the best-known algorithm for solving SRFPs.","bibtex":"@Article{2009passch,\n\tauthor = {R. Pasupathy and B. W. Schmeiser},\n\ttitle = {Retrospective-approximation algorithms for multidimensional stochastic root-finding problems},\n\tjournal = {ACM TOMACS},\n\tyear = {2009},\n\tvolume = {19},\n\tnumber = {2},\n\tpages = {5:1--5:36},\n\tdoi = {10.1145/1502787.1502788},\n\turl = {http://web.ics.purdue.edu/~pasupath/PAPERS/2009passch.pdf},\n\tkeywords = {retrospective approximation, stochastic root-finding},\n\tabstract = {The stochastic root-finding problem (SRFP) is that of solving a nonlinear system of equations using only a simulation that provides estimates of the functions at requested points. Equivalently, SRFPs seek locations where an unknown vector function attains a given target using only a simulation capable of providing estimates of the function. SRFPs find application in a wide variety of physical settings. We develop a family of retrospective-approximation (RA) algorithms called Bounding RA that efficiently solves a certain class of multidimensional SRFPs. During each iteration, Bounding RA generates and solves a sample-path problem by identifying a polytope of stipulated diameter, with an image that bounds the given target to within stipulated tolerance. Across iterations, the stipulations become increasingly stringent, resulting in a sequence of shrinking polytopes that approach the correct solution. Efficiency results from: (i) the RA structure, (ii) the idea of using bounding polytopes to exploit problem structure, and (iii) careful step-size and direction choice during algorithm evolution. Bounding RA has good finite-time performance that is robust with respect to the location of the initial solution, and algorithm parameter values. Empirical tests suggest that Bounding RA outperforms Simultaneous Perturbation Stochastic Approximation (SPSA), which is arguably the best-known algorithm for solving SRFPs.}}\n\n\n\n\n%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%\n%%% PRELIMINARY BOOK CITATIONS\n\n","author_short":["Pasupathy, R.","Schmeiser, B. W."],"key":"2009passch","id":"2009passch","bibbaseid":"pasupathy-schmeiser-retrospectiveapproximationalgorithmsformultidimensionalstochasticrootfindingproblems-2009","role":"author","urls":{"Paper":"http://web.ics.purdue.edu/~pasupath/PAPERS/2009passch.pdf"},"keyword":["retrospective approximation","stochastic root-finding"],"metadata":{"authorlinks":{}},"downloads":0,"html":""},"bibtype":"article","biburl":"web.ics.purdue.edu/~pasupath/rpVitapublist.bib","dataSources":["Q4TNB6PfH2Ne5F2fj"],"keywords":["retrospective approximation","stochastic root-finding"],"search_terms":["retrospective","approximation","algorithms","multidimensional","stochastic","root","finding","problems","pasupathy","schmeiser"],"title":"Retrospective-approximation algorithms for multidimensional stochastic root-finding problems","year":2009}