Skip to content
ML Atlas

LeNet-5

Yann LeCun's classic convolutional network for handwritten digits, which established the CNN blueprint and helped read millions of bank cheques a day in the 1990s.

Also LeNet · cheque reading · MNIST

Fig. / How it works
IMAGEFEATURE MAPA 3×3 FILTER SLIDES OVER THE IMAGE
Role · kind
Real systemsFamous system
Level
Beginner

LeCun, Bottou, Bengio and Haffner, AT&T Bell Labs, 1998

One of the first computer brains that could reliably read handwritten numbers, like the amount on a cheque, by looking for small strokes and building them up into digits.

Where it fits

Computer VisionComputer VisionConvolutional Neural NetworkConvolutional Neural NetworkBackpropagationBackpropagationLeNet-5
+All relationships as a list

Uses / built on

See it on the full map

How it works

Long before deep learning was a buzzword, a team at AT&T Bell Labs led by Yann LeCun built neural networks that read handwriting. Their 1998 paper described LeNet-5 and showed that a convolutional neural network trained end to end with backpropagation beat other approaches at recognizing digits.

The architecture already contains the CNN recipe:

  1. Convolution: small learned filters slide over the image to detect strokes and edges.
  2. Subsampling (pooling): shrink the feature maps, keeping the important signals.
  3. Repeat, then a few fully connected layers decide which digit it is.

The paper also described a complete cheque-reading system based on these networks, deployed commercially and reading several million cheques per day. And it introduced MNIST, the dataset of handwritten digits that became the “hello world” of machine learning.

Fourteen years later, AlexNet scaled the same blueprint up with GPUs and ImageNet, and started the deep learning boom. See also OCR.

Everyday use cases

  • Case 01

    How you've probably met it

    Indirectly, if you deposited a cheque in the 1990s, and directly as the first CNN in most deep learning courses, usually trained on MNIST digits.

Strengths, weaknesses and when to use it

Strengths

  • Compact, effective design that proved learned features beat hand-designed ones for handwriting.

Weaknesses

  • Far too small for natural photos; deep CNNs needed GPUs and big datasets (AlexNet, 2012).

When to use itAs the historical starting point for CNNs and a first hands-on exercise.

Compare with

Go deeper

Sources

  1. [1]LeCun et al., Gradient-Based Learning Applied to Document Recognition, Proceedings of the IEEE (1998)
  2. [2]Wikipedia: MNIST database

Facts checked as of 2026-10-12