Home | Events | Calibration of agent-based models with a black-box optimization approach par Jean-Daniel Kant
Calibration of agent-based models with a black-box optimization approach par Jean-Daniel Kant
AFD - 75012
15 March 2019 to
15 March 2019
03:00 PM - 05:00 PM
The next session of the Research Seminar of the Chair Energy and Prosperity will take place on Friday 15 of March at 15:00 PM at the Agence Française de Développement. We will receive ean-Daniel KANT (Sorbonne Université Sciences, LIP6) for an initial speach on:
Calibration of agent-based models with a black-box optimization approach
ABSTRACT
The validation of agent-based models (ABM) in a crucial issue to help the development and dissemination of ABMs in economics and social sciences. After recalling the problematics of such a validation, I will present a typology of common approaches, and then describe the method we used at LIP6 to calibrate an ABMs of the French Labor Market : WorkSim.
WorkSim is a big model, and has 60 parameters to calibrate, so this is a quite challenging problem. We propose to use a black-box optimization approach, where an evolutionary algorithm, CMA-ES, is used to minimize a fitness function that aggregates the distances between (60) simulated outputs and parameter targets. I will present and discuss our calibration results, show the advantages and limits of such an approach and extend the discussion to what still has to be done to improve ABMs validation.
General information Date : March 15, 2019 from 03:00 PM to 05:00PM Place : AFD – 5 Rue Roland Barthes, 75012 Paris – Room E04 100
Attention : une pièce d’identité vous sera demandée à l’entrée du bâtiment et prévoyez d’être un peu en avance (du fait des contrôles de sécurité).
