Joel Dyer

Career Profile

I am a senior postdoctoral researcher at the University of Oxford’s Department of Computer Science and a Senior Research Fellow at the Oxford Institute for New Economic Thinking. I draw on computational statistics and machine learning to develop tools that help decision-makers reason about and experiment with complex systems using (often agent-based) simulation models.

I completed a DPhil in computational statistics and machine learning at the University of Oxford’s Mathematical Institute and Institute for New Economic Thinking under the supervision of Prof. J. Doyne Farmer in 2022. During this time, I held visiting positions at The Alan Turing Institute and the University of Bristol as an Alan Turing Enrichment Student, and afterwards worked as a Postdoctoral Research Scientist at Improbable before returning to academia.

Experiences

Senior Postdoctoral Researcher

Present

University of Oxford

Senior Research Fellow

Present

Institute for New Economic Thinking, Oxford

Research Scientist

2022 - 2023

Improbable

Data Science Researcher

Summer 2021

Energy Systems Catapult

Videos/projects/blogs/talks

Some videos, projects, and blogs can be found at the following links:

ICAIF 2024 Tutorial on Differentiable Agent-based Modelling - We delivered a tutorial on building and calibrating differentiable agent-based models at ICAIF 2024

AAMAS 2024 Tutorial on Differentiable Agent-based Modelling - We delivered a tutorial on building and calibrating differentiable agent-based models at AAMAS 2024

Gradient-assisted calibration for (differentiable and non-differentiable) financial agent-based models - Slides from my ICAIF '23 talk on Bayesian calibration of differentiable agent-based models in finance

Variational Bayesian inference for (differentiable and non-differentiable) agent-based models - Recording of my INET Oxford talk on the use of differentiable programming and generalised variational Bayesian inference to calibrate both differentiable and (standard) non-differentiable agent-based models. Slides available here

Black-box Bayesian inference for agent-based models - Recording of my INET Oxford talk on Black-box Bayesian inference for agent-based models

Calibrating agent-based models to microdata with graph neural networks - Recording of my AI4ABM talk on our work on Calibrating agent-based models to microdata with graph neural networks

The expected largest degree in a random attachment model of network growth - Mathematics and simulation study of the expected largest degree in a random attachment model of network growth

Amortised likelihood-free inference for expensive time-series simulators with signatured density ratio estimation - Recording of my AISTATS 2022 talk on our work on Amortised likelihood-free inference for expensive time-series simulators with signatured density ratio estimation. A longer version is available here

Blog post on my role as a Data Science Challenge Facilitator at one of The Alan Turing Institute's Data Study Groups - Associated report is here

Approximate Bayesian computation with path signatures - Recording of my One World ABC talk on our work on Approximate Bayesian computation with path signatures

Software

Some of my open-source software projects include:

sbi4abm - A toolbox consisting of flexible, simulation-efficient calibration methods for agent-based models

synthpop - A toolbox accompanying our AAMAS 2024 paper on scenario generation through population synthesis in agent-based models

BlackBIRDS - A toolbox consisting of methods to perform Black-Box Inference foR Differentiable Simulators. Published in The Journal of Open Source Software

SignatuRE - Code for our AISTATS 2022 paper 'Amortised likelihood-free inference for expensive time-series simulators with signatured ratio estimation'