RMC Workshop/Seminar: Parallel Computing in R + ML in Agent Based Modelling

About the speaker: George G. Vega Yon is an Assistant Professor of Research at the Division of Epidemiology at the University of Utah, United States, and a Data Scientist at the US Centers for Disease Control and Prevention. He works on studying Complex Systems using Statistical Computing. He has over ten years of experience developing scientific software focusing on high-performance computing, statistical analysis, and social network analysis. Dr. Vega Yon is a methodologist working across fields, including sociology, epidemiology, and genetics. His training is in Public Policy (M.A. Universidad Adolfo Ibáñez, Chile), Economics (M.Sc. California Institute of Technology, USA), and Biostatistics (Ph.D. University of Southern California, USA). You can learn more about George on his website: https://ggvy.cl.

Talk: Machine Learning Sandwich for Agent-Based Models: Automatic calibration and post-processing adjustment for ABMs

Agent-based models [ABMs] are flexible computational tools used to simulate complex systems. With ABMs, we can approximate challenging questions for which analytical solutions are elusive. Two challenges persist in the field: model calibration and accuracy of its predictions. Calibration, which entails adjusting model parameters to match observed data, is too often computationally expensive with no clear general solution. Moreover, the problem becomes more pressing in situations involving policy decisions requiring prompt action. On the other hand, although, compared to other models, ABMs can embed more complicated dynamics, it generally comes with the cost of reduced accuracy. This presentation will introduce a general framework for addressing these issues. This framework, which we call "machine learning [ML] sandwich for ABMs," advances the field by embedding ML into agent-based modeling as a pre and post-processing layer. Furthermore, the post-processing layer is an application of Mechanistic Machine Learning [MechML,] a hybrid between theory and data-driven prediction methods that have been demonstrated to boost prediction accuracy.

Workshop: "High-performance computing for the layman" or "How did I stop worrying and sped up my code": A lightning workshop with an application in R

This 1-hour long lightning workshop will introduce the attendees to general concepts of parallel computing and how to leverage it in using the R programming language. The first half will cover high-level concepts and essential considerations when working with parallel computing (including pseudo-random number generation and parallelization strategies). The second part will introduce attendees to the R package parallel using examples for numerical approximations and simulations. Registering participants will have access to a pre-configured R environment on posit.cloud.