Efficient Optimization Algorithms for Parameter Selection of Agent-Based Models

Agent-based models (ABMs) are powerful computational tools used to simulate and study complex systems in various fields such as social science, epidemiology, and economics. These models rely on individual agents and their interactions to capture emergent phenomena that cannot be understood through traditional analytical approaches. However, the effectiveness and validity of ABMs depend heavily on selecting appropriate model parameters as nonlinear inverse problems—a challenging task due to the high-dimensional, non-linear, and often stochastic nature of these models.

The proposed PhD project aims to develop efficient optimization algorithms for parameter selection in ABMs [1], via designing tailored sketching schemes [2,3] for dimensionality reduction, and Learning-to-Optimize schemes [4] which leverage deep neural networks to learn the structure of the underlying optimization problem for numerical acceleration.