Patrick Cannon


Scholar      GitHub      LinkedIn      Twitter

Hi there! I’m Patrick. Welcome to my personal website.

I’m a machine learning researcher, and currently lead the research and engineering function of a computer vision startup I co-founded in early 2023.

Previously, I worked at Improbable where I focused on simulation-based inference methods (like normalising flows) for calibrating complex scientific simulators, including agent-based models.

I have a PhD in statistics from the University of Bristol, where I worked on Markov chain Monte Carlo techniques for inference in coalescent population genetic models, supervised by Christophe Andrieu and Mark Beaumont.

Broadly, my interests lie in probabilistic machine learning and its use in solving important technological and scientific challenges. Guided by a statistical worldview, I try to leverage the tools of uncertainty quantification to improve our understanding of black-box algorithms, particularly deep learning. I’m mostly concerned with:

🔘   Robust deep learning
🔘   Algorithmic trustworthiness
🔘   Alignment

I believe these are some of the most pressing challenges in artificial intelligence and, as a result, are not only technically interesting but highly consequential for humanity.


For a full list, please see my google scholar page.

Robust Neural Posterior Estimation and Statistical Model Criticism
Daniel Ward, Patrick Cannon, Mark Beaumont, Matteo Fasiolo, Sebastian Schmon
NeurIPS, 2022

Investigating the Impact of Model Misspecification in Simulation-Based Inference
Patrick Cannon, Daniel Ward, Sebastian Schmon
arXiv preprint, 2022

Amortised Inference for Expensive Time-Series Simulators with Signatured Ratio Estimation
Joel Dyer, Patrick Cannon, Sebastian Schmon

Calibrating Agent-Based Models to Microdata with Graph Neural Networks
Joel Dyer, Patrick Cannon, J. Doyne Farmer, Sebastian Schmon
ICML, AI4ABM Workshop, 2022. Spotlight.

High Performance Simulation for Scalable Multi-Agent Reinforcement Learning
Jordan Langham-Lopez, Patrick Cannon, Sebastian Schmon
ICML, AI4ABM Workshop, 2022. Spotlight.

Generalized Posteriors in Approximate Bayesian Computation
Sebastian Schmon*, Patrick Cannon*, Jeremias Knoblauch
AABI (Symposium on Advances in Approximate Bayesian Inference), 2021

Approximate Bayesian Computation with Path Signatures
Joel Dyer, Patrick Cannon, Sebastian Schmon
arXiv preprint, 2021

Deep Signature Statistics for Likelihood-free Time-series Models
Joel Dyer, Patrick Cannon, Sebastian Schmon
ICML, INNF+ Workshop, 2021