Congratulations to INET Oxford doctoral students Joel Dyer and Aymeric Vié for both winning awards at the AI4ABM Workshop in July.
This was the inaugural workshop focusing on artificial intelligence methods in and for agent-based modelling at the July 2022 International Conference on Machine Learning. Its purpose was to stimulate and share existing research into the use of recent developments in AI and to support progress in agent-based modelling, such as methods for parameter inference, model validation, and analysing and accelerating ABMs. It brought together an interdisciplinary group of attendees from across academia and industry to deliver a range of talks and poster sessions.
Joel Dyer was awarded Best Proposal Paper for 'Calibrating agent-based models to microdata using graph neural networks'. This paper concerns how (dynamic) graph neural networks can be employed in neural simulation-based inference procedures (such as for the problem of estimating parameters) to naturally capture the (dynamic) graph structure of agent-based models when there is fine, granular data available on the system being modeled.
Aymeric Vié was awarded Best Paper in the Benchmark category for 'Evology: an empirically-calibrated market ecology agent-based model for trading strategy search'. This paper presents the next iteration of the market ecology model previously published by Scholl, Calinescu and Farmer. By creating a financial agent-based model with a variety of trading strategies, the authors hope to understand market ecology dynamics, namely how the composition of the stock market changes and how different traders interact. This article highlights several model improvements regarding realistic investment flows, heterogeneity in trading strategies, and profit-driven adaptation of fund behaviour.