Yann LeCun is executive chairman of AMI Labs and a professor at New York University. His research connects neural-network image recognition with systems that learn representations and models of the world.[10][11]
Start with Gradient Based Learning Applied to Document Recognition →
Education
DEA in artificial intelligence and pattern recognition
Completed 1984
[3]
Links
X Personal website NYU Executive chairman · AMI Labs
Jacob T. Schwartz Professor · New York University
Background
LeCun’s education spans electrical engineering and computer science. NYU records his engineering diploma from ESIEE in 1983, a DEA in artificial intelligence and pattern recognition in 1984, and a computer-science doctorate from Université Pierre et Marie Curie in 1987. Those subjects recur in his work: learning systems must connect mathematical objectives with the variability of real images and signals.[3]
Career
LeCun now identifies himself as AMI Labs’ executive chairman while continuing as a professor at NYU. AMI’s own team page lists him as chairman. The company describes its program as building systems that understand the physical world, retain memories, reason and plan. These are research objectives. They explain the focus of his current organization without claiming that human-level machine intelligence has been achieved. He was Meta’s chief AI scientist from 2018 to 2025, after serving as the founding director of Meta-FAIR from 2013 to 2017.[2][10][11]
Contributions
Learning directly from document images
The 1998 paper Gradient Based Learning Applied to Document Recognition, coauthored with Léon Bottou, Yoshua Bengio and Patrick Haffner, explains how neural networks can learn to recognize handwritten characters from pixel images. Its broader argument is that useful features can be learned instead of entirely specified by hand. It also treats document recognition as a system of interacting modules, including segmentation and language modeling, and describes training them against an overall objective. This makes it a substantial account of both image recognition and the engineering of complete recognition systems.[4][9]
Predicting representations of missing information
LeCun is a coauthor of I-JEPA, a 2023 method for learning from images without supplied category labels. The system sees a context region and predicts representations of other regions in the same image. A representation is a learned numerical description of content; the task is not to reconstruct every missing pixel. The paper’s masking strategy encourages useful information about objects and scenes. Its experiments evaluate the resulting representations on downstream tasks, rather than demonstrating an autonomous agent that can plan in the physical world.[5]
Tools, writing & talks
Writing, research & talks
Start here
A detailed coauthored account of learned handwriting recognition and training complete document-processing systems. The introduction is a useful entry point; the mathematical and architectural sections reward familiarity with neural networks.
The introduction explains the practical problem of recognition before the technical details of learning systems.
Coauthor[4]The I-JEPA paper turns a representation-prediction idea into a specific image-learning experiment. Read it to see the relationship between context, masking and learned visual features, with measured results limited to its evaluation tasks.
Coauthor[5]- A Path Towards Autonomous Machine Intelligenceopenreview.net
LeCun’s 2022 position paper sets out a proposed route toward autonomous machine intelligence. It is a research agenda to compare with concrete experiments such as I-JEPA, rather than a report that the entire proposed system has been built.
Author[6]