Léon Bottou
2007 Regional Award Winner — Faculty
Current Position:
Principal Researcher
Institution:
Facebook AI Research NYC (Previously at Microsoft Research and NEC Labs America)
Discipline:
Computer Science
Recognized for: Developing DjVu document compression method; his research work on stochastic gradient algorithms for large scale machine learning
Areas of Research Interest and Expertise: Large-scale machine learning, statistical machine learning, structured learning systems, compound image compression
Biography:
- PhD, Computer Science, Université de Paris XI, France
- MSc, Applied and Fundamental Mathematics and Computer Science, l'École Normale Supérieure, France
- MSc, Engineering, l'École Polytechnique, France
Léon Bottou’s long-term interests are in both large-scale learning and large-scale inference. In his work, he is using a pragmatic strategy that he learned while working withLarry Jackel at the AT&T Bell Labs in the nineties: “Find the most difficult problem you can solve today and push the boundary. When every known technique breaks downs, you have an opportunity to do good science.” The long term goal of Dr. Bottou’s research is to understand how to build machines with human-level intelligence.
Léon Bottou played a major role in the development of an image pattern recognition system that incorporated novel methods to enable simultaneous segmenting and recognition of images of strings of sloppily handwritten characters. His DjVu document compression method uses highly efficient algorithms to segment and compress bi-tonal and color images, photos, or text documents into foreground and background components.
Since receiving the Blavatnik Award, Dr. Battou’s research has evolved along two lines: (1) understanding how abundant training data changes the nature of the learning problem, and (2) understanding the interaction of learning and reasoning. Both directions found a good application in a recent work applying counterfactual reasoning to the machine learning problems associated with ad placement on the web.
"The long term goal of my research is to understand how to build machines with human-level intelligence. Reaching this goal will require important conceptual advances that I cannot honestly anticipate at this point. However I am convinced that these missing concepts will teach us a lot about ourselves and about our learning and reasoning abilities.”
Key Publications:
- Bottou L, Peters J, Quiñonero-Candela J, Denis X, Charles DX, Chickering M, Portugualy E, Ray D, Simard P, Snelson E. Counterfactual Reasoning and Learning Systems, 2012.
- Bottou L, Bousquet O. The Tradeoffs of Large Scale Learning, Advances in Neural Information Processing Systems, Edited by J.C. Platt, D. Koller, Y. Singer and S. Roweis, NIPS Foundation, 2008.
- Bottou L, Haffner P, Howard PG, Simard P, Bengio Y, Cun YL. High Quality Document Image Compression with DjVu, Journal of Electronic Imaging, 1998.