[01125] Generalized Optimization Algorithms for $M$-Estimation of Complex Simulation Models
Session Time & Room : 3E (Aug.23, 17:40-19:20) @E503
Type : Contributed Talk
Abstract : We provide a new optimization algorithm for simulation-based models with highly irregular objective functions, like those of network or agent-based models. The approximate inexact Newton method (AINM) is based on approximating the first two derivatives of the function through a polynomial regression. We provide new general results concerning approximate Netwon methods, we extensively discuss the theoretical and computational aspects of the AINM, and support the theory by Monte Carlo experiments and two applications.
