The History of AI

DanNet

Era 3 · From the expert boom to the statistical turn (1980–2011) › Hardware and data

Lineage
Machines and computation
Kind
Implementation

Who

IDSIA

What happened

DanNet, a convolutional network trained on GPUs by Dan Cireşan, Ueli Meier, Jonathan Masci, and Jürgen Schmidhuber at the Swiss lab IDSIA, achieves in 2011 the first better-than-human result in a visual recognition competition, one on traffic signs.

Why it mattered

The first GPU-trained convolutional network to reach superhuman performance in a vision competition, a year before AlexNet won ImageNet with a similar recipe.

The full story

Convolutional networks had existed since the late 1980s, when LeCun used them to read handwritten digits. They had the right structure for images, but training them at scale was too slow, and for years other methods dominated computer vision.

Schmidhuber's group at IDSIA wrote a fast GPU implementation of convolutional networks. With it they could train networks deeper and wider than usual, with max-pooling layers, and combine several separately trained networks into a multi-column system whose answers were averaged.

In 2011, in a traffic sign recognition competition, DanNet's accuracy beat that of the people who took the same test. It was the first time a machine had done so in a visual recognition competition. The group published the method in the journal Neural Networks in 2012.

The result showed that the combination of deep convolutional networks, GPUs, and large datasets already beat hand-designed vision methods. A year later AlexNet applied a very similar recipe to ImageNet and won over the rest of the field.

Builds on

  • LeNet-5 and MNIST (1998)

    Documented direct influence

    DanNet is a LeNet-style convolutional network, deeper and trained on GPUs; the paper cites the 1998 work by LeCun et al.

Leads to

  • AlexNet (September 30, 2012)

    Documented direct influence

    The AlexNet paper cites two papers by Cireşan et al., including the one on multi-column networks, and says its architecture is "somewhat similar" to the "columnar" CNN of Cireşan et al. (section 3.2)

Threads that pass through here

Sources

Original source

How to cite

“DanNet.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/dannet

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline