Jon McAuliffe

Also known as Jon D. McAuliffe

Jon McAuliffe is a statistician whose machine learning research includes supervised topic models, variational inference, and methods for analyzing astronomical images. He studied computer science and applied mathematics at Harvard and completed a statistics PhD at UC Berkeley in 2005 with Michael Jordan. His industry work included recommender systems at Amazon, and he joined Wharton’s Statistics Department in 2005. With David Blei, he introduced supervised topic models in 2007; his later work included a 2017 review of variational inference and the 2023 StarNet method for separating overlapping light sources in astronomical images. As of October 2026, he is an Adjunct Professor in Berkeley’s Department of Statistics and a cofounder and Co-Chief Investment Officer of Voleon.

Now at: University of California, Berkeley and The Voleon Group
Works: 45
Citations: 6,250
h-index: 18
Badges: Founder
Bridge

Publication numbers from OpenAlex, as of 28 October 2026.

Timeline

1995–1999

Trading systems at Shaw

From 1995 to 1999, he works at D. E. Shaw & Co., where he develops statistical equity arbitrage trading systems.
2005

Harvard degrees and Berkeley doctorate

He holds an AB in computer science with honors and an SM in applied mathematics from Harvard University. He completes a statistics PhD at UC Berkeley in 2005, advised by Michael Jordan. His earlier industry work includes recommender systems at Amazon.com, click prediction at Efficient Frontier, and genotype-calling algorithms at Affymetrix.
2005

The Erich Lehmann Award

In 2005, he wins the Berkeley Statistics Department’s Erich Lehmann Award for an outstanding PhD dissertation in theoretical statistics.
2005

Assistant professor at Wharton

In 2005, he joins the University of Pennsylvania’s Wharton School as an assistant professor in the Statistics Department. His research there includes comparative genomics and bioinformatics.
2007

Supervised topic models

In 2007, he and David M. Blei introduce supervised latent Dirichlet allocation in “Supervised Topic Models.” The model learns from labeled documents to predict responses, with tests on movie ratings from reviews and web page popularity from text descriptions.
2017

A variational inference review

In 2017, he coauthors “Variational Inference: A Review for Statisticians” with David M. Blei and Alp Kucukelbir in the Journal of the American Statistical Association. The article explains how optimization can approximate probability distributions and describes stochastic optimization for large datasets.
2017

The TrustVI algorithm

In 2017, he coauthors “Fast Black-box Variational Inference through Stochastic Trust-Region Optimization” with Jeffrey Regier and Michael I. Jordan. They introduce TrustVI, an algorithm for variational inference, and implement it in the Stan framework.
2019

The Bloomberg 50

In 2019, he is included in “The Bloomberg 50: The People Who Changed Global Business in 2019.”
2023

StarNet separates overlapping stars

In 2023, he coauthors “Variational Inference for Deblending Crowded Starfields” with Runjing Liu, Jeffrey Regier, and the LSST Dark Energy Science Collaboration in the Journal of Machine Learning Research. Their StarNet method uses Bayesian inference to distinguish individual light sources in crowded astronomical images.
As of October 2026

Berkeley professor and Voleon cofounder

As of October 2026, he is an Adjunct Professor in UC Berkeley’s Department of Statistics and a cofounder and Co-Chief Investment Officer of Voleon. His Berkeley research interests include machine learning, statistical prediction, variational inference, and reinforcement learning.