R-MGSPLINE: Retrospective multi-gradient search for multi-objective simulation optimization on integer lattices

R-MGSPLINE: Retrospective multi-gradient search for multi-objective simulation optimization on integer lattices. Applegate, E. A. & Hunter, S. R. In Mustafee, N., Bae, K. G., Lazarova-Molnar, S., Rabe, M., Szabo, C., Haas, P., & Son, Y., editors, Proceedings of the 2019 Winter Simulation Conference, pages 3516–3527, Piscataway, NJ, 2019. IEEE.

Paper doi abstract bibtex

We introduce the Retrospective Multi-Gradient Search with Piecewise Linear Interpolation and Neighborhood Enumeration (R-MGSPLINE) algorithm for finding a local efficient point when solving a multi-objective simulation optimization problem on an integer lattice. In this nonlinear optimization problem, all objectives can only be observed with stochastic error, the decision variables are integer-valued, and a local solution is called the efficient set. R-MGSPLINE uses a retrospective approximation (RA) framework to repeatedly call the MGSPLINE sample-path solver at a sequence of increasing sample sizes, using the solution from the previous RA iteration as a warm start for the current RA iteration. The MGSPLINE algorithm performs a line search along a common descent direction constructed from pseudo-gradients of each objective, followed by a neighborhood enumeration for certification. Numerical experiments show that R-MGSPLINE converges to a local weakly efficient point.

@inproceedings{2019apphunWSC,
	Year = {2019},
	Author = {E. A. Applegate and S. R. Hunter},
	Title = {{R-MGSPLINE}: {R}etrospective multi-gradient search for multi-objective simulation optimization on integer lattices},
	Booktitle = {Proceedings of the 2019 Winter Simulation Conference},
	Editor = {N. Mustafee and {K.-H.} G. Bae and S. {Lazarova-Molnar} and M. Rabe and C. Szabo and P. Haas and {Y.-J.} Son},
	Publisher = {IEEE},
         Address = {Piscataway, NJ},
	doi = {10.1109/WSC40007.2019.9004719},
	pages = {3516--3527},
	url_Paper = {http://web.ics.purdue.edu/~hunter63/PAPERS/2019apphunWSC.pdf},
	abstract = {We introduce the Retrospective Multi-Gradient Search with Piecewise Linear Interpolation and Neighborhood Enumeration (R-MGSPLINE) algorithm for finding a local efficient point when solving a multi-objective simulation optimization problem on an integer lattice.  In this nonlinear optimization problem, all objectives can only be observed with stochastic error, the decision variables are integer-valued, and a local solution is called the efficient set.  R-MGSPLINE uses a retrospective approximation (RA) framework to repeatedly call the MGSPLINE sample-path solver at a sequence of increasing sample sizes, using the solution from the previous RA iteration as a warm start for the current RA iteration.  The MGSPLINE algorithm performs a line search along a common descent direction constructed from pseudo-gradients of each objective, followed by a neighborhood enumeration for certification. Numerical experiments show that R-MGSPLINE converges to a local weakly efficient point.},
	keywords = {simulation optimization > multi-objective > integer-ordered}}

Downloads: 0