John Michael Winn

Also known as John Winn, J. Winn, John M. Winn

John Winn is a machine learning researcher at Microsoft Research Cambridge whose work spans probabilistic inference, computer vision and healthcare applications. After studying at Cambridge under Chris Bishop and David MacKay, he coauthored Variational Message Passing with Bishop in 2005 and the TextonBoost segmentation model in 2006. With Tom Minka, he devised Infer.NET, a library for inference in graphical models that won a Patents for Humanity award in 2013, and he coauthored the Pascal VOC benchmark paper in 2009. His book Model-Based Machine Learning appeared in print in 2023. He is a Senior Principal Researcher in the Machine Learning and Perception group at Microsoft Research Cambridge.

Now at: Microsoft Research

Timeline

2005

Variational Message Passing

After an undergraduate degree in Electrical and Information Sciences at Cambridge University and PhD studies in the Inference Group at the Cavendish Laboratory under Chris Bishop and David MacKay, he coauthors Variational Message Passing with Bishop in 2005. The paper in the Journal of Machine Learning Research introduces a general-purpose algorithm for variational inference in Bayesian networks, implemented in an engine called VIBES.
2006

TextonBoost for image segmentation

In 2006, he coauthors TextonBoost with Jamie Shotton, Carsten Rother and Antonio Criminisi at the European Conference on Computer Vision. The model uses texton-based features that jointly model shape and texture, with shared boosting for classification and a conditional random field for segmenting photographs.
2008

Infer.NET with Tom Minka

In February 2008, he presents Infer.NET in a seminar at the Gatsby Computational Neuroscience Unit in London. The Microsoft Research project, joint work with Tom Minka, aims to create a general-purpose software library for inference in graphical models that runs message-passing algorithms efficiently on large-scale models.
2009

The Pascal VOC benchmark

In 2009, he coauthors The Pascal Visual Object Classes (VOC) Challenge in the International Journal of Computer Vision with Mark Everingham, Luc Van Gool, Christopher K. I. Williams and Andrew Zisserman. The paper describes the dataset and evaluation procedure of the annual challenge, which has run since 2005 and has become accepted as the benchmark for object detection.
2010

Probabilistic programming with Infer.NET

In June 2010, he gives a public talk at Microsoft Research Cambridge on Infer.NET and probabilistic programming. He has worked on Infer.NET since its inception, with applications from recognising Bill Gates' glasses in live video to understanding the genetic causes of asthma.
2013

A Patents for Humanity award

In April 2013, Infer.NET wins the information technology category of the first Patents for Humanity awards from the US Patent and Trademark Office. The library's inventors include him and Tom Minka, and he takes a particular interest in medical and biological applications, including gene analysis with the Wellcome Trust Sanger Institute.
2023

The Model-Based Machine Learning book

In 2023, his book Model-Based Machine Learning appears in print from Taylor & Francis. He writes it with Christopher M. Bishop, Thomas Diethe, John Guiver and Yordan Zaykov, and each chapter works through one real-world case study.
As of October 2026

Senior Principal Researcher at Microsoft

He is a Senior Principal Researcher at Microsoft Research Cambridge, in the Machine Learning and Perception group. His main research interests are machine learning, machine vision, ML for healthcare and information extraction.