Yann LeCun

Yann LeCun (born 1960) is the French-born computer scientist who developed convolutional neural networks (CNNs). LeNet (1989), which he built at Bell Labs to read handwritten digits, went on to read bank checks and became the prototype of the CNNs that took over image recognition in the 2010s. He shared the 2018 ACM Turing Award with Geoffrey Hinton and Yoshua Bengio, and afterwards, from the head of Meta's AI research lab FAIR, kept arguing against the field's turn to large language models alone. In 2025 he left Meta to start AMI Labs, a company building world models.

Background

According to ACM's laureate biography, LeCun grew up in the suburbs of Paris, where his father, an engineer, passed on an interest in electronics and mechanics. His name was originally written Le Cun; he dropped the space after finding that Americans took "Le" for a middle name. He earned an engineering diploma from ESIEE in Paris in 1983, began independent research on machine learning as a student, and took a PhD in computer science at the Université Pierre et Marie Curie (now Sorbonne Université) in 1987.

ACM notes that his research ran parallel to, and independent of, Hinton's. Both were drawn to the then-unfashionable neural-network approach, and both found that what was later called backpropagation could train hidden layers and overcome the limits of simple networks. A 1985 workshop at Les Houches in the French Alps put him in touch with the international community; a few months later Hinton, visiting Paris, introduced himself, which led to a postdoctoral year in Hinton's group at Toronto in 1987-88.

Bell Labs: CNNs and checks

His most far-reaching work at Bell Labs was the convolutional network. A simple network trained to respond to a feature in the top left of an image will not respond to the same feature elsewhere; a convolutional network sweeps a small filter across the whole input, so higher layers notice a pattern wherever it occurs. Training became faster and networks smaller. ACM's award announcement says that in the late 1980s, at Toronto and Bell Labs, LeCun was the first to train a CNN on images of handwritten digits.

The papers, datasets, accuracy figures and zip-code reading are on the LeNet page. According to ACM, the check-reading system he built in the early 1990s with Bengio, Léon Bottou and Patrick Haffner was licensed to bank-system suppliers such as NCR, and LeCun suggests that at one point it was reading more than 10% of all checks written in the US. Another contribution from these years was "Optimal Brain Damage", a way of shrinking networks by removing unnecessary connections.

When AT&T spun off most of Bell Labs as Lucent Technologies in 1996, LeCun stayed at AT&T to run an image-processing group, where with Bottou, Haffner and Paul G. Howard he developed the DjVu compression technology. It compressed scanned documents more effectively than Adobe Acrobat and was used extensively by the Internet Archive in the early 2000s.

The Turing Award (2018)

On 27 March 2019 ACM named Bengio, Hinton and LeCun recipients of the 2018 Turing Award "for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing". It honoured the three as among the small group that remained committed to neural networks by the early 2000s, and listed three contributions by LeCun:

• Convolutional neural networks: developed in the 1980s and now standard in computer vision, speech recognition, image synthesis and more
• Improving backpropagation: he proposed an early version, gave a clean derivation based on variational principles, and described two simple methods to speed up learning (see backpropagation)
• Broadening the vision of neural networks: with Bottou, he proposed the idea, used in every modern deep-learning framework, of building learning systems as networks of modules trained by backpropagation through automatic differentiation

The three are also linked through CIFAR's Learning in Machines & Brains programme, which LeCun has co-directed with Bengio since 2014. For the revival of deep learning around 2006, see deep belief nets.

FAIR and Meta (2013-2025)

According to ACM, as the big technology companies competed for neural-network researchers in the early 2010s, Facebook hired LeCun in December 2013 to create FAIR, which he led from New York until 2018 while keeping his NYU post. He became the public face of AI at Facebook, regularly in newspapers and magazines. In 2018 he stepped down as director to become Chief AI Scientist, focusing on strategy and scientific leadership. He has consistently championed open-source development and open research (The Decoder); during his tenure as Chief AI Scientist, in 2023, Meta released the large language model LLaMA to researchers.

In these years LeCun also consistently dismissed the argument that AI threatens humanity's survival. The contrast with Hinton, his co-laureate, when Hinton left Google in 2023 to speak about AI risk is drawn on that event's page.

Leaving Meta and AMI Labs (2025-)

In November 2025 LeCun announced that he would leave Meta at the end of the year and pursue Advanced Machine Intelligence (AMI) research in a new company. According to The Decoder, he described his twelve years at Meta as five as founding director of FAIR and seven as Chief AI Scientist, and called the creation of FAIR his proudest non-technical achievement. The new company would aim at systems that understand the physical world, have persistent memory, and can reason and plan complex actions, with Meta as a partner. Coverage cited as background Meta's shift toward large language models and product-driven work, and the waning of his influence after the creation of a superintelligence lab and the appointment of a new chief AI officer.

In March 2026 the Paris-headquartered AMI Labs announced a seed round of about $1.03 billion (€890 million, per heise). Tech.eu reported a $3.5 billion valuation and called it Europe's largest-ever seed round. Alexandre LeBrun, formerly chief executive of the medical-AI company Nabla, is CEO; LeCun is chair; and the company has sites in Paris, New York, Montreal and Singapore. Its programme — world models that understand physical reality by predicting how the world behaves, rather than large language models — is the one LeCun had argued for since his FAIR years.

The arc

LeCun's career tracks one idea becoming an industry standard over thirty years. The architecture that read zip codes in 1989 became, by way of AlexNet in 2012, the mainstream of image recognition — and at the height of that success he moved to the other side, arguing that language models, the next presumed mainstream, would not by themselves reach intelligence. Founding his own company in his mid-sixties, after twelve years leading research inside a corporation, is a continuation of that argument.