CLE Seminar: Aldo Glielmo
Seminar Series
Friday, 13 December 2024, 12 p.m.
Room G.122
Aldo GLIELMO, Banca d'Italia
“Machine Learning for Economic ABMs: From Fast Calibration to AI-Agents”
Abstract: From their revolutionary applications first in computer science and then in physics, chemistry or biology, Artificial Intelligence (AI) and Machine Learning (ML) are now driving a paradigm shift in economic research, through their capacity to automatically improve simply via larger datasets and increased computational power. Agent-Based Models (ABMs) are uniquely positioned to lead this AI-driven transformation in economic modelling, thanks to their inherently computational nature. First, I will present two state-of-the-art and open-source software packages: ABCredIT and BeforeIT, designed to provide robust and extensible modelling frameworks to facilitate ABM usage and experimentations. Second, calibrating ABM parameters on large datasets is becoming increasingly feasible through ML schemes. I will discuss Black-IT, a dedicated calibration engine for ABMs, and explore how Reinforcement Learning (RL) techniques can enhance the computational efficiency of the calibration process. Finally, the integration of AI-software agents in the ABM methodology could eventually relieve modellers from the need to explicitly define by hand complex behavioural rules. I will illustrate the effects of integrating RL agents into traditional ABMs and discuss the promises and challenges of incorporating agents that are based on Large Language Models (LLMs) in the not-too-distant future.