1965–, Computer scientist
Léon Bottou is a French computer scientist whose long collaboration with Yann LeCun produced foundational work on stochastic gradient descent for large-scale learning. The 1998 paper Online Learning and Stochastic Approximations and the 2008 paper The Tradeoffs of Large Scale Learning (with Olivier Bousquet) gave SGD the modern theoretical analysis that justified its use at scale in deep learning.
Bottou co-developed DjVu, the document-image compression format that long preceded modern PDF for scientific archives, and the Lush scientific computing language. He has been a major contributor to LeNet (with LeCun) and to graph transformer networks for handwriting recognition. He spent much of his career at AT&T, NEC Labs, Microsoft Research and now Facebook AI Research.
Video
Related people: Yann LeCun
Works cited in this book:
- Gradient-based learning applied to document recognition (1998) (with Y. Lecun, Y. Bengio, P. Haffner)
Discussed in:
- Chapter 10: Training & Optimisation, Training Optimisation