WIZER: Automated model improvement in multi-agent social-network systems

Abstract

There has been a significant increase in the use of multi-agent social-network models due to their ability to flexibly model emergent behaviors in complex socio-technical Systems while linking to real data. These models are growing in size and complexity which requires significant time and effort to calibrate, validate, improve the model, and gain insight into model behavior. In this paper, we present our knowledge-based simulation-aided approach for automating model-improvement and our tool implementing this approach (WIZER). WIZER is capable of calibrating and validating multi-agent social-network models, and facilitates model-improvement and understanding. By employing knowledge-based search, causal analysis, and simulation control and inference techniques, WIZER can reduce the number of simulation runs needed to calibrate, validate, and improve a model and improve the focus of these runs. WIZER automates reasoning and analysis of simulations, instead of being a multi-agent programming language or environment. We ran a preliminary version of WIZER on BioWar a city-scale social agent network Simulation of the effects of weaponized biological attacks on a demographically-realistic population within a background of naturally-occurring diseases. The results demonstrate the efficacy of WIZER. © 2006 Springer Science+Business Media, Inc.