Olivier Chapelle

Olivier Chapelle was a machine learning researcher whose work covered support vector machines, semi-supervised learning, web-search ranking and online advertising. With Vladimir Vapnik he published on choosing support vector machine parameters in 1999 and 2002, and in 2006 he co-edited the MIT Press volume Semi-Supervised Learning with Bernhard Schölkopf and Alexander Zien. As a Senior Research Scientist in Machine Learning at Yahoo, he organized the Learning to Rank Challenge with Yi Chang in 2010, and in 2011 he and Lihong Li published an empirical evaluation of Thompson sampling. At Criteo he worked on conversion modeling and was Principal Research Scientist there in 2015, and by 2020 he worked at Google. He died in 2020 at age 42, and NeurIPS held a memorial session in his honour.

Timeline

1999–2002

Tuning support vector machines

In 1999, he and Vladimir Vapnik introduce new functionals for choosing support vector machine parameters, based on the span of support vectors and rescaling of the feature space. In 2002, he, Vapnik, Olivier Bousquet and Sayan Mukherjee publish "Choosing Multiple Parameters for Support Vector Machines" in Machine Learning, which tunes several parameters at once by gradient descent on estimates of the generalization error.
2006–2007

Editing Semi-Supervised Learning

He co-edits Semi-Supervised Learning with Bernhard Schölkopf and Alexander Zien, published by MIT Press in 2006 as a comprehensive review of using unlabeled data in classification problems. In 2007, he is credited with Léon Bottou and Dennis DeCoste on Large-Scale Kernel Machines.
2010–2011

The Yahoo! ranking challenge

A Senior Research Scientist in Machine Learning at Yahoo, he organizes the Learning to Rank Challenge with Yi Chang in spring 2010. The two write its overview paper, published in 2011.
2011

Thompson sampling put to the test

In 2011, he and Lihong Li present "An Empirical Evaluation of Thompson Sampling" at NIPS, showing on simulated and real data that the heuristic is highly competitive. They argue that it should be part of the standard baselines to compare against.
2014–2015

Conversion modeling at Criteo

In August 2014, Criteo's AI Lab announces that he will present his paper on conversion modeling at KDD and releases the associated dataset of conversion logs. His paper "Simple and Scalable Response Prediction for Display Advertising" appears in ACM Transactions on Intelligent Systems and Technology in 2014, and in 2015 a Criteo press release names him Principal Research Scientist at Criteo.
By 2020

A researcher at Google

By 2020, he works at Google on machine learning.
2020

A NeurIPS memorial session

He dies in 2020 at age 42, after an unexpected severe illness. NeurIPS holds a memorial session in his honour, with contributions from friends including Olivier Bousquet, Bernhard Schölkopf and Vladimir Vapnik.