Yann LeCun

вики
, , , ,
1

French-American computer scientist; inventor of convolutional neural networks, Turing Award laureate, and pioneer of the world-model paradigm for machine intelligence.

Basic Information / Profile

Field Details
Full Name Yann André Le Cun
Born July 8, 1960, Soisy-sous-Montmorency, France
Nationality French-American
Current Institutions AMI Labs (Executive Chair); New York University (Jacob T. Schwartz Chaired Professor)
Research Fields Deep learning, convolutional neural networks, computer vision, energy-based models, world models, image compression, robotics
PhD Advisor Maurice Milgram
PhD Thesis Modèles connexionnistes de l’apprentissage (Connectionist Learning Models), Université Pierre et Marie Curie, 1987
Personal Website yann.lecun.com
X / Twitter @ylecun
GitHub github.com/ylecun
Google Scholar scholar.google.com/citations?user=WLN3QrAAAAAJ
NYU Faculty Page engineering.nyu.edu/faculty/yann-lecun
ACM Turing Award amturing.acm.org/award_winners/lecun_6017366.cfm

Overview

Yann LeCun is a French-American computer scientist widely acknowledged as the primary architect of convolutional neural networks (CNNs) — the class of models that transformed computer vision and drove the commercial deployment of deep learning in the 2010s. Trained in France and shaped by a postdoctoral year under Geoffrey Hinton at Toronto, he developed LeNet at Bell Labs in the late 1980s and 1990s, a network that was deployed at scale to read handwritten cheques across the United States and whose architecture underpins virtually every visual recognition system in use today. In 2018 he shared the Turing Award with Hinton and Yoshua Bengio — the trio are commonly referred to as the “Godfathers of Deep Learning.” He spent twelve years as Chief AI Scientist at Meta, where he led the Fundamental AI Research (FAIR) laboratory and championed open-source AI development. In November 2025 he left Meta to found Advanced Machine Intelligence Labs (AMI Labs), a venture focused on world-model architectures, which raised $1.03 billion in March 2026. He holds the Jacob T. Schwartz Chaired Professorship at NYU’s Courant Institute, a position he has maintained alongside his industry roles since 2003.

Early Life & Education

Childhood and Formation in France

LeCun was born on July 8, 1960, in Soisy-sous-Montmorency, a suburb north of Paris. His surname derives from the old Breton form Le Cunff, with roots in the Guingamp region of Brittany; his given name Yann is the Breton equivalent of Jean. He grew up in France during a period when connectionism — the idea that cognition could be modelled by networks of simple units — was a minority position in AI, and he developed an early interest in the theoretical possibility of learning machines. By his late twenties he was publishing proposals for learning algorithms that would take decades to fully vindicate.

ESIEE Paris (1980–1983)

LeCun received a Diplôme d’Ingénieur from ESIEE Paris in 1983, an engineering qualification in the French grande école tradition. The programme gave him a rigorous grounding in mathematics and signal processing that would later inform his approach to neural network architecture.

Université Pierre et Marie Curie — PhD (1983–1987)

He completed his doctorate in computer science at the Université Pierre et Marie Curie (now Sorbonne University) in 1987, supervised by Maurice Milgram. His thesis, Modèles connexionnistes de l’apprentissage, proposed an early form of the backpropagation algorithm — independently and nearly simultaneously with David Rumelhart, Geoffrey Hinton, and Ronald Williams — and applied it to multilayer networks. A 1985 paper at Cognitiva described an early asymmetric-threshold network learning scheme, predating the canonical 1986 Rumelhart–Hinton–Williams publication in Nature.

Postdoctoral Research under Hinton, University of Toronto (1987–1988)

Before joining AT&T, LeCun spent a year as a postdoctoral researcher at the University of Toronto under Geoffrey Hinton, where he deepened his understanding of neural network training dynamics and began the intellectual collaboration that would shape the field’s development. This year cemented the working relationship between LeCun, Hinton, and the broader Toronto group.

Career

AT&T Bell Laboratories — Research Scientist (1988–1996)

In 1988 LeCun joined the Adaptive Systems Research Department at AT&T Bell Labs in Holmdel, New Jersey, led by Lawrence Jackel. Over eight years he produced the work that defined his scientific legacy. His 1989 paper “Backpropagation Applied to Handwritten Zip Code Recognition” introduced the convolutional neural network architecture in its mature form, combining local receptive fields, weight sharing, and spatial subsampling in a biologically motivated hierarchy. By 1998 this culminated in LeNet-5, presented in “Gradient-Based Learning Applied to Document Recognition” (Proceedings of the IEEE, 1998, with Léon Bottou, Yoshua Bengio, and Patrick Haffner), a network trained end-to-end to read handwritten digits. The associated bank cheque recognition system was deployed by NCR and others, reading roughly 10% of all cheques written in the United States at its peak. He also developed Optimal Brain Damage (NeurIPS 1990), a principled pruning technique for removing redundant weights — a precursor to the model compression methods now central to efficient inference — and Graph Transformer Networks for sequence-labelling tasks.

AT&T Labs-Research — Head of Image Processing (1996–2003)

After the Bell Labs breakup, LeCun moved to AT&T Labs-Research as head of the Image Processing Research Department. The primary output of this period was DjVu, a document image compression format developed with Léon Bottou and Patrick Haffner, designed for efficient web distribution of scanned documents and later adopted by the Internet Archive. He also co-developed the Lush programming language with Bottou and collaborated with Vladimir Vapnik, the inventor of support vector machines. After a brief fellowship at NEC Research Institute, he joined New York University in 2003.

New York University — Jacob T. Schwartz Chaired Professor (2003–present)

At NYU’s Courant Institute of Mathematical Sciences, LeCun holds joint appointments in Computer Science, Data Science, Neural Science, and Electrical and Computer Engineering. His NYU research focused on energy-based models (EBMs) — a framework for unsupervised and self-supervised learning that treats inference as an optimisation problem over an energy landscape rather than a probabilistic normalisation — and on feature learning for multi-stage object recognition architectures. He co-directed the NYU machine learning group with Rob Fergus and others, training a generation of researchers including Raia Hadsell, Marc’Aurelio Ranzato, and Koray Kavukcuoglu. In 2012 he became the founding director of the NYU Center for Data Science, stepping down in 2014 when his Meta role expanded. In 2013, with Yoshua Bengio, he co-founded the International Conference on Learning Representations (ICLR), which grew to become one of the field’s premier venues. In 2016 he was Visiting Professor at the Collège de France, where he delivered the inaugural lecture for the Chaire Informatique et Sciences Numériques. In 2023 he was named inaugural holder of the Jacob T. Schwartz Chair at the Courant Institute.

Meta (Facebook) AI — VP and Chief AI Scientist / FAIR Director (2013–2025)

On December 9, 2013, LeCun became the first Director of Facebook AI Research (FAIR) in New York City, a role that evolved into VP and Chief AI Scientist as Meta grew. He built FAIR into one of the most productive academic-style industrial AI laboratories, with a philosophy of open publication and open-source release: FAIR was instrumental in the development and open-sourcing of PyTorch (the framework that came to dominate deep learning research), as well as large-scale computer vision models, fairness and robustness benchmarks, and self-supervised learning methods. During his tenure he sustained his contrarian public stance: he argued persistently against AI existential risk narratives, challenged the framing of large language models as a path to general intelligence, and maintained that current AI systems lack the world-understanding necessary for robust real-world action. On November 19, 2025, LeCun confirmed his departure from Meta after twelve years to found his own company.

AMI Labs (Advanced Machine Intelligence Labs) — Executive Chair (2025–present)

LeCun co-founded Advanced Machine Intelligence Labs (AMI Labs) in late 2025, with Alex LeBrun serving as CEO and LeCun as Executive Chair. The company’s research thesis is that large language models represent a structural dead end for achieving human-level machine intelligence, and that the correct path is the construction of world models — predictive systems that learn the structure and dynamics of the physical world from multi-modal sensory input, enabling safe planning and action rather than pattern-matched text generation. In March 2026, AMI announced a $1.03 billion funding round co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, at a $3.5 billion pre-money valuation. In January 2026, LeCun also became founding Chair of the Technical Research Board of Logical Intelligence, an AI company developing energy-based reasoning systems.

Key Contributions

  • Convolutional Neural Networks / LeNet — Developed the CNN architecture at Bell Labs between 1989 and 1998, combining biologically inspired local connectivity, weight sharing, and pooling in a hierarchy trained end-to-end by backpropagation; the paper “Gradient-Based Learning Applied to Document Recognition” (Proceedings of the IEEE, 1998) remains one of the most cited works in computer science.
  • Backpropagation — early independent formulation — Proposed a practical backpropagation learning algorithm in his 1987 PhD thesis, independently of and concurrently with the canonical Rumelhart–Hinton–Williams formulation.
  • Optimal Brain Damage (NeurIPS 1990) — Introduced a principled second-order method for network weight pruning based on computing the Hessian of the loss; a foundational contribution to model compression and efficient inference.
  • Bank Cheque OCR System — Applied LeNet in a commercial system deployed by NCR that processed approximately 10% of US bank cheques at peak; one of the earliest large-scale industrial deployments of a neural network.
  • DjVu Image Compression (1998, with Léon Bottou and Patrick Haffner) — Designed a document image compression format optimised for scanned text that achieved significantly better compression than JPEG at equivalent quality and was adopted by the Internet Archive.
  • Energy-Based Models — Developed a unified theoretical framework for supervised, unsupervised, and self-supervised learning based on energy minimisation rather than probabilistic normalisation, articulated in a 2006 tutorial and subsequent work; influenced the design of modern contrastive and joint-embedding self-supervised methods.
  • Graph Transformer Networks — Developed sequence-labelling architectures combining convolutional features with structured prediction, applied to handwriting recognition and OCR.
  • ICLR (co-founded 2013, with Yoshua Bengio) — Established the International Conference on Learning Representations with an open peer review process, now one of the most competitive and influential AI conferences.
  • FAIR and PyTorch — Led Meta’s Fundamental AI Research lab and oversaw or facilitated the open-source release of PyTorch, which displaced TensorFlow as the dominant framework in academic deep learning research.
  • World Models / AMI Labs — Articulated and is actively building toward a paradigm shift from language-model pre-training to architecture that learns a predictive model of physical reality from multi-modal inputs, enabling safe planning and action in the world.

Awards & Recognition

  • ACM Turing Award (2018, jointly with Geoffrey Hinton and Yoshua Bengio) — Computing’s highest honour, shared with the two other “Godfathers of Deep Learning” for conceptual and engineering breakthroughs in neural networks.
  • IEEE Neural Network Pioneer Award (2014)
  • PAMI Distinguished Researcher Award (2015)
  • IRI Medal (2018)
  • Harold Pender Award, University of Pennsylvania (2018)
  • Fellow, AAAI (2019)
  • Member, US National Academy of Sciences (2021)
  • Member, US National Academy of Engineering
  • Member, French Académie des Sciences
  • Princess of Asturias Award for Scientific Research (2022, jointly with Hinton, Bengio, and Demis Hassabis)
  • Chevalier de la Légion d’Honneur (2023) — Awarded by the President of France.
  • VinFuture Grand Prize (2024, jointly with Bengio, Hinton, Jensen Huang, and Fei-Fei Li)
  • Queen Elizabeth Prize for Engineering (2025, jointly with Bengio, Bill Dally, Hinton, John Hopfield, Jensen Huang, and Fei-Fei Li) — Engineering’s most prestigious international prize.
  • Honorary Doctorates from IPN Mexico City (2016), EPFL (2018), Université Côte d’Azur (2021), Università di Siena (2023), Hong Kong University of Science and Technology (2023).

Key Relationships

  • Geoffrey Hinton — Postdoctoral mentor and intellectual forebear; co-Turing Award winner; the two share both credit and occasional public disagreement, particularly on AI safety timelines, where LeCun argues Hinton’s alarm is overstated.
  • Yoshua Bengio — Co-Turing Award winner and co-founder of ICLR; collaborated on the landmark LeNet paper and on the foundational 2015 Nature survey “Deep learning”; the two share a long history of parallel and complementary research.
  • Léon Bottou — The most sustained of LeCun’s research collaborations, spanning Bell Labs, AT&T, and FAIR; co-inventor of DjVu, the Lush language, and stochastic gradient descent analysis work.
  • Vladimir Vapnik — Colleague at AT&T Labs and inventor of support vector machines; intellectual interlocutor during the period when kernel methods and neural networks competed for primacy.
  • Raia Hadsell — NYU doctoral student who became a leading researcher at DeepMind; her work on autonomous off-road driving with LeCun was an early demonstration of learned perception for mobile robotics.
  • Koray Kavukcuoglu — NYU and FAIR collaborator; later became VP of Research at DeepMind; co-authored work on multi-stage object recognition architectures.
  • Mark Zuckerberg — Recruited LeCun to build FAIR in 2013; their twelve-year working relationship gave LeCun unusual latitude to pursue fundamental research within a product-driven company.
  • Alex LeBrun — CEO of AMI Labs, the company LeCun founded in 2025; leads the operational side while LeCun serves as Executive Chair and chief research strategist.

Personal Style

LeCun is one of the most publicly combative figures in AI — prolific on social media, willing to name opposing views as wrong, and unconcerned with diplomatic hedging when he believes the field is heading in a mistaken direction. His rejection of AI doomism is categorical and consistent: he has argued for years that fears of near-term superintelligent risk are misguided and that current systems, including large language models, lack the structural properties needed for human-level intelligence. His April 2026 Brown University lecture captured the register well — telling a standing-room audience that “AI sucks” and that LLM-centric investment is “complete BS” — positions he has held in substance since before they became fashionable contrarianism. Technically, he is defined by architectural thinking: his most durable contributions (CNNs, energy-based models, world models) are not individual algorithms but frameworks that reorganise how a class of problems should be approached. He has been a consistent advocate for open science, pushing for open publication at FAIR when competitors kept research proprietary, and his Breton origins surface in a mild pride: he signs his name Yann rather than Jean and notes the etymology when asked.

References