TY - GEN
T1 - Alternate social theory discovery using genetic programming
T2 - 2017 Genetic and Evolutionary Computation Conference, GECCO 2017
AU - Gunaratne, Chathika
AU - Garibay, Ivan
N1 - Publisher Copyright:
© 2017 ACM.
PY - 2017/7/1
Y1 - 2017/7/1
N2 - A pressing issue with agent-based model (ABM) replicability is the ambiguity behind micro-behavior rules of the agents. In practice, modelers choose between competing theories, each describing separate candidate solutions. Pattern-oriented modeling (POM) and stylized facts matching recommend testing theories against patterns extracted from real-world data. Yet, manually, POM is tedious and prone to human error. In this study, we present a genetic programming strategy to evolve debatable assumptions on agent micro-behaviors. After proper modularization of the candidate micro-behaviors, genetic programming can discover candidate micro-behaviors which reproduce patterns found in real-world data. We illustrate this strategy by evolving the decision tree representing the farm-seeking strategy of agents in the Artificial Anasazi ABM. Through evolutionary theory discovery, we obtain multiple candidate decision trees for farm-seeking which fit the archaeological data better than the calibrated original model in the literature. We emphasize the necessity to explore a range of components that influence the agents' decision making process and demonstrate that this is achievable through an evolutionary process if the rules are modularized as required. The end result is a set of plausible candidate solutions that closely fit the real-world data, which can then be nominated by domain experts.
AB - A pressing issue with agent-based model (ABM) replicability is the ambiguity behind micro-behavior rules of the agents. In practice, modelers choose between competing theories, each describing separate candidate solutions. Pattern-oriented modeling (POM) and stylized facts matching recommend testing theories against patterns extracted from real-world data. Yet, manually, POM is tedious and prone to human error. In this study, we present a genetic programming strategy to evolve debatable assumptions on agent micro-behaviors. After proper modularization of the candidate micro-behaviors, genetic programming can discover candidate micro-behaviors which reproduce patterns found in real-world data. We illustrate this strategy by evolving the decision tree representing the farm-seeking strategy of agents in the Artificial Anasazi ABM. Through evolutionary theory discovery, we obtain multiple candidate decision trees for farm-seeking which fit the archaeological data better than the calibrated original model in the literature. We emphasize the necessity to explore a range of components that influence the agents' decision making process and demonstrate that this is achievable through an evolutionary process if the rules are modularized as required. The end result is a set of plausible candidate solutions that closely fit the real-world data, which can then be nominated by domain experts.
KW - Agent-based modeling
KW - Artificial Anasazi
KW - Calibration
KW - Genetic programming
KW - Theory discovery
UR - http://www.scopus.com/inward/record.url?scp=85026370436&partnerID=8YFLogxK
U2 - 10.1145/3071178.3071332
DO - 10.1145/3071178.3071332
M3 - Conference contribution
AN - SCOPUS:85026370436
T3 - GECCO 2017 - Proceedings of the 2017 Genetic and Evolutionary Computation Conference
SP - 115
EP - 122
BT - GECCO 2017 - Proceedings of the 2017 Genetic and Evolutionary Computation Conference
PB - Association for Computing Machinery, Inc
Y2 - 15 July 2017 through 19 July 2017
ER -