Léon Bottou
Machine learning researcher
A machine learning researcher whose work established stochastic gradient descent as a core learning algorithm and who co-created DjVu.
Biography
Léon Bottou is a French computer scientist best known for his work in machine learning and data compression, which presented stochastic gradient descent as a fundamental learning algorithm. Born in 1965, he earned the Diplôme d'Ingénieur from École Polytechnique in 1987, a Magistère from École Normale Supérieure in 1988 and a PhD from Université Paris-Sud in 1991.
He collaborated with Yann LeCun on SN, a neural network simulation package, and worked at AT&T Bell Laboratories on handwriting and check recognition systems that read over 10% of US checks in the late 1990s and early 2000s. He is one of the main creators of the DjVu image compression technology and the original developer of the Lush programming language. He worked at NEC Laboratories from 2002 to 2010, then at Microsoft, and joined Facebook Artificial Intelligence Research in 2015 as a research lead. He was program chair of NeurIPS in 2013 and ICML in 2009.
Notable contributions
Established stochastic gradient descent for large-scale learning
His work argued that stochastic gradient descent reaches similar loss to batch methods with the same number of samples but runs faster on large datasets, making it a fundamental learning algorithm.
Co-created DjVu
With Yann LeCun and Patrick Haffner, created the DjVu image compression technology, used by sites such as the Internet Archive to distribute scanned documents.
Built check-reading systems at Bell Labs
Helped develop a bank check recognition system deployed by NCR and others that read over 10% of checks in the US in the late 1990s and early 2000s.
Career history
Principal researcher, Microsoft Research New York
