Synthetic Populations

All projectsResearch · 2026

Create synthetic individuals from aggregate statistics while preserving population constraints.

Maximum entropy · Aggregate statistics

Imagine All The People: a diverse illustrated population

Overview

Synthetic populations are collections of artificial individuals constructed from aggregate statistics. My research explores how explicit constraints and maximum-entropy models can reproduce relationships among attributes, providing populations for simulation and analysis.

Visit ImagineAllThePeople · Read the research paper

Research approach

A population model must account for combinations of attributes as well as their individual frequencies. The work with Jean-Daniel Zucker studies a maximum-entropy relaxation of multi-way cardinality constraints, making it possible to fit a probabilistic model without enumerating every possible attribute combination.

The theoretical model matches target frequencies in expectation. Its practical implementation uses an approximate MCMC fitting procedure; the paper evaluates the resulting populations on survey-derived and independently generated benchmarks.

Paper and PDFs

Maximum Entropy Relaxation of Multi-Way Cardinality Constraints for Synthetic Population Generation
François Pachet and Jean-Daniel Zucker
Frontiers in Artificial Intelligence, 9, article 1840945, 23 October 2026.
Published paper (PDF) · DOI
Earlier arXiv preprint · Preprint PDF

ImagineAllThePeople

This research is connected to my work at ImagineAllThePeople. The company develops synthetic populations and simulation tools for exploring how people and societies may respond to decisions. Its website presents the approach and applications.

Back to Projects · Browse Publications