ABM calibration allows model parameters to be tuned to fit an observed dataset, however what if we could go one step further and learn the symbolic code of the model itself? In this research project we demonstrate a new genetic programming approach which is able to evolve interpretable logic for the agent update function of an ABM from scratch. We employ a flexible domain-specific language (DSL) which consists of basic mathematical building blocks. The flexibility of our method is demonstrated by learning symbolic models in two different domains: bird flocking and opinion dynamics, targeting data produced by hand-written reference models. We show that the evolved solutions are behaviourally identical to the reference models and generalise extremely well.
Agent-based models are a particularly promising application of genetic programming, since relatively simple behavioural rules at the agent level can lead to complex behaviour at the macro level. This can be considered a form of Inverse Generative Social Science, where searching over the space of possible models can provide new understanding of phenomena observed in real world data.
Rory Greig Bio:
Rory Greig is a senior research scientist in the defence division at Improbable, with a background in software engineering building large scale simulations. He is working on tools for modellers to build and calibrate high performance ABMs.