← All research spotlights

Research spotlight · Unit 9

Yann LeCun

New York University

Research field
Convolutional networks, machine learning, computer vision

Yann LeCun built the first convolutional networks trained by backpropagation at Bell Laboratories from 1988 on, and the 1998 paper with Léon Bottou, Yoshua Bengio, and Patrick Haffner described their mature form, LeNet-5, together with the MNIST database that the field has used ever since. The paper asked whether a network could learn its own feature detectors from the pixels, trained end to end by gradient descent, and answered with a test error of 0.95 per cent on handwritten digits and a cheque-reading system that was deployed commercially and read several million cheques a day. LeCun shared the 2018 Turing Award with Bengio and Geoffrey Hinton. The unit presents the paper after its layer-by-layer account of LeNet-5, as the origin of every network in the unit.

In the masterclass

Section 4, after the layers and the parameter count of LeNet-5, presents the 1998 paper and its cheque-reading system.

The research is presented in Unit 9: Convolutional Networks and Computer Vision, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.

The paper

LeCun, Y., Bottou, L., Bengio, Y., Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278‑2324. https://doi.org/10.1109/5.726791

The scientists named here are not contributors to the AI Masterclass, are not affiliated with it, and have not endorsed it. The program presents and explains their published research, with references to the original papers.