The Connectionist Revival

Yann LeCun

Invented the convolutional neural network in the late 1980s and deployed it commercially reading handwritten checks years before deep learning was fashionable.

Yann LeCun (b. 1960) developed the convolutional neural network architecture while working with Geoffrey Hinton at Bell Labs in the late 1980s, building on earlier work on the neocognitron. His key idea was weight sharing: instead of learning a separate weight for every pixel position, a small kernel slides across the whole image, so the network learns the same edge- or texture-detector regardless of where it appears in frame. That single design choice made image recognition tractable with far fewer parameters than a fully connected network would need.

Unlike most of the field, LeCun's early CNN work wasn't just academic — his LeNet architecture was deployed commercially by banks in the 1990s to read handwritten digits on checks, reportedly processing a meaningful share of checks in the US at the time. That made LeCun one of the few researchers with a working, revenue-generating deep learning system a full two decades before AlexNet made the approach fashionable again.

He later became the founding director of Facebook AI Research (FAIR) and shared the 2018 Turing Award with Hinton and Yoshua Bengio. He has been a vocal skeptic of claims that scaling large language models alone will reach general intelligence, arguing instead for architectures with an explicit world model — a live disagreement with parts of the field pursuing pure scale.

See also: Geoffrey Hinton, Yoshua Bengio

Learn more: Computer Vision · Wikipedia: Yann LeCun

Mentioned in

Lessons where this comes up in context.