Agent-Based Model Exploration and Calibration using Machine Learning Surrogates. Lamperti, F., Roventini, A., & Sani, A. Working Paper LEM 2017-11, 2017. ![Agent-Based Model Exploration and Calibration using Machine Learning Surrogates [pdf]](https://bibbase.org/img/filetypes/pdf.svg "pdf")
Website abstract bibtex Bringing Agent-Based Models closer to the data is an open challenge. While facilitating the comparison to more standard approaches, getting closer to the data promotes Agent-Based Models as a methodology. In this paper, we explicitly tackle parameter space exploration and the issue of calibrating agent-based models using real data. Traditionally, three computationally expensive steps are involved: running the Agent-Based Model, measuring the calibration quality to real data and locating the parameters of interest. We demonstrate dramatic improvements in computation time by replacing the expensive Agent-Based Model and calibration measure with a machine learning surrogate that approximates a fast approximation of 1st and total order sensitivities over the parameter space. Our approach facilitates parameter exploration by a policy-maker, while providing a powerful filter to gain intuition and insight into Agent-Based Models with relatively large parameter spaces. We illustrate our approach by means of two agent-based models calibrated against different objective functions and evaluated on a large out-of-sample sensitivity analysis.
@article{
title = {Agent-Based Model Exploration and Calibration using Machine Learning Surrogates},
type = {article},
year = {2017},
websites = {http://www.lem.sssup.it/WPLem/files/2017-11.pdf},
id = {df9b1313-a55f-3596-aafd-71ec250c5d5f},
created = {2016-10-31T11:02:53.000Z},
file_attached = {false},
profile_id = {8856f0a8-e934-3b54-83b3-1dc0f138e432},
group_id = {bd0f1172-1660-3fe3-b7d0-9f2d0b4051ad},
last_modified = {2018-06-12T14:42:17.580Z},
tags = {DOLFINS_T1.3,DOLFINS_T2.3,DOLFINS_WP1,DOLFINS_WP2},
read = {false},
starred = {false},
authored = {false},
confirmed = {true},
hidden = {false},
citation_key = {Lamperti2016},
private_publication = {false},
abstract = {Bringing Agent-Based Models closer to the data is an open challenge. While facilitating the comparison to more standard approaches, getting closer to the data promotes Agent-Based Models as a methodology. In this paper, we explicitly tackle parameter space exploration and the issue of calibrating agent-based models using real data. Traditionally, three computationally expensive steps are involved: running the Agent-Based Model, measuring the calibration quality to real data and locating the parameters of interest. We demonstrate dramatic improvements in computation time by replacing the expensive Agent-Based Model and calibration measure with a machine learning surrogate that approximates a fast approximation of 1st and total order sensitivities over the parameter space. Our approach facilitates parameter exploration by a policy-maker, while providing a powerful filter to gain intuition and insight into Agent-Based Models with relatively large parameter spaces. We illustrate our approach by means of two agent-based models calibrated against different objective functions and evaluated on a large out-of-sample sensitivity analysis.},
bibtype = {article},
author = {Lamperti, Francesco and Roventini, Andrea and Sani, Amir},
journal = {Working Paper LEM 2017-11}
}
Downloads: 0
{"_id":"2ocJBmyCkDpFpWtr7","bibbaseid":"lamperti-roventini-sani-agentbasedmodelexplorationandcalibrationusingmachinelearningsurrogates-2017","downloads":0,"creationDate":"2018-06-19T08:32:54.822Z","title":"Agent-Based Model Exploration and Calibration using Machine Learning Surrogates","author_short":["Lamperti, F.","Roventini, A.","Sani, A."],"year":2017,"bibtype":"article","biburl":null,"bibdata":{"title":"Agent-Based Model Exploration and Calibration using Machine Learning Surrogates","type":"article","year":"2017","websites":"http://www.lem.sssup.it/WPLem/files/2017-11.pdf","id":"df9b1313-a55f-3596-aafd-71ec250c5d5f","created":"2016-10-31T11:02:53.000Z","file_attached":false,"profile_id":"8856f0a8-e934-3b54-83b3-1dc0f138e432","group_id":"bd0f1172-1660-3fe3-b7d0-9f2d0b4051ad","last_modified":"2018-06-12T14:42:17.580Z","tags":"DOLFINS_T1.3,DOLFINS_T2.3,DOLFINS_WP1,DOLFINS_WP2","read":false,"starred":false,"authored":false,"confirmed":"true","hidden":false,"citation_key":"Lamperti2016","private_publication":false,"abstract":"Bringing Agent-Based Models closer to the data is an open challenge. While facilitating the comparison to more standard approaches, getting closer to the data promotes Agent-Based Models as a methodology. In this paper, we explicitly tackle parameter space exploration and the issue of calibrating agent-based models using real data. Traditionally, three computationally expensive steps are involved: running the Agent-Based Model, measuring the calibration quality to real data and locating the parameters of interest. We demonstrate dramatic improvements in computation time by replacing the expensive Agent-Based Model and calibration measure with a machine learning surrogate that approximates a fast approximation of 1st and total order sensitivities over the parameter space. Our approach facilitates parameter exploration by a policy-maker, while providing a powerful filter to gain intuition and insight into Agent-Based Models with relatively large parameter spaces. We illustrate our approach by means of two agent-based models calibrated against different objective functions and evaluated on a large out-of-sample sensitivity analysis.","bibtype":"article","author":"Lamperti, Francesco and Roventini, Andrea and Sani, Amir","journal":"Working Paper LEM 2017-11","bibtex":"@article{\n title = {Agent-Based Model Exploration and Calibration using Machine Learning Surrogates},\n type = {article},\n year = {2017},\n websites = {http://www.lem.sssup.it/WPLem/files/2017-11.pdf},\n id = {df9b1313-a55f-3596-aafd-71ec250c5d5f},\n created = {2016-10-31T11:02:53.000Z},\n file_attached = {false},\n profile_id = {8856f0a8-e934-3b54-83b3-1dc0f138e432},\n group_id = {bd0f1172-1660-3fe3-b7d0-9f2d0b4051ad},\n last_modified = {2018-06-12T14:42:17.580Z},\n tags = {DOLFINS_T1.3,DOLFINS_T2.3,DOLFINS_WP1,DOLFINS_WP2},\n read = {false},\n starred = {false},\n authored = {false},\n confirmed = {true},\n hidden = {false},\n citation_key = {Lamperti2016},\n private_publication = {false},\n abstract = {Bringing Agent-Based Models closer to the data is an open challenge. While facilitating the comparison to more standard approaches, getting closer to the data promotes Agent-Based Models as a methodology. In this paper, we explicitly tackle parameter space exploration and the issue of calibrating agent-based models using real data. Traditionally, three computationally expensive steps are involved: running the Agent-Based Model, measuring the calibration quality to real data and locating the parameters of interest. We demonstrate dramatic improvements in computation time by replacing the expensive Agent-Based Model and calibration measure with a machine learning surrogate that approximates a fast approximation of 1st and total order sensitivities over the parameter space. Our approach facilitates parameter exploration by a policy-maker, while providing a powerful filter to gain intuition and insight into Agent-Based Models with relatively large parameter spaces. We illustrate our approach by means of two agent-based models calibrated against different objective functions and evaluated on a large out-of-sample sensitivity analysis.},\n bibtype = {article},\n author = {Lamperti, Francesco and Roventini, Andrea and Sani, Amir},\n journal = {Working Paper LEM 2017-11}\n}","author_short":["Lamperti, F.","Roventini, A.","Sani, A."],"urls":{"Website":"http://www.lem.sssup.it/WPLem/files/2017-11.pdf"},"bibbaseid":"lamperti-roventini-sani-agentbasedmodelexplorationandcalibrationusingmachinelearningsurrogates-2017","role":"author","downloads":0},"search_terms":["agent","based","model","exploration","calibration","using","machine","learning","surrogates","lamperti","roventini","sani"],"keywords":[],"authorIDs":[]}