Yann LeCun Patents & Innovation Profile — PatSnap Eureka
Yann LeCun: Patent Portfolio & Innovation Analysis
Yann LeCun is a French-American computer scientist and Turing Award laureate who holds 159 patents spanning convolutional neural networks, optical character recognition, biometric authentication, and computer vision, with filings from 1989 to 2021. His portfolio is primarily assigned to Element, Inc. and AT&T Bell Laboratories, and covers foundational methods that underpin modern deep learning and mobile identity systems.
Patent Filing Activity
Peak years were 1993 and 2018, each with 16 filings — reflecting AT&T's character recognition commercialisation and Element's global biometric rollout.
Yann LeCun's Patent Filing Patterns
Three distinct career phases are visible in the timeline: a prolific Bell Laboratories era peaking in 1993, a brief NEC phase in 2005, and an Element, Inc. biometrics phase peaking in 2018.
Annual Patent Filings
Peak years were 1993 (16 filings, AT&T character recognition) and 2018 (16 filings, Element biometric global rollout).
Technology Domain Breakdown
Image recognition (G06V30) is the largest single domain with 24 patents, followed by face detection/biometrics (G06K9) with 13.
Core Areas of Innovation
Yann LeCun's portfolio spans four decades of applied neural network research, from foundational character recognition to cutting-edge mobile biometric authentication.
Image Recognition & Optical Character Recognition
24 patentsLeCun's earliest and most technically foundational work centres on machine reading of handwritten and printed characters using convolutional neural networks. These patents address reliable recognition despite variations in handwriting style, scale, slant, and noise, built to serve postal automation and banking cheque processing systems.
- Hierarchical constrained automatic learning neural network for character recognition (US5067164A)
- Time delay neural network for printed and cursive handwritten character recognition (US5105468A)
- Hierarchical constrained automatic learning network for character recognition (US5058179A)
Biometric Authentication & Anti-Spoofing
10 patentsThe largest bloc of LeCun's recent patent activity covers contactless palm-print biometric authentication using camera-equipped mobile devices, and methods for detecting spoofing of 3D objects using 2D representations. Anti-spoofing patents introduce dynamic authentication patterns displayed on device screens to detect fraudulent 2D representations of real faces or palms.
- System and method for biometric authentication in connection with camera equipped devices (US20140068740A1)
- Methods, systems, and media for detecting spoofing in mobile authentication (US11425562B2)
- System and method for biometric authentication — EP (EP2893489B1)
Face Detection & Pose Estimation
13 patentsDuring his time at NEC Laboratories America, LeCun co-invented methods for simultaneous face detection and pose estimation using energy-based convolutional neural networks. These systems are robust to variations in skin colour, facial hair, eyeglasses, scale, and lighting, training a CNN to map face images to points on a pose manifold.
- Synergistic face detection and pose estimation with energy-based models (US7236615B2)
- Synergistic face detection and pose estimation with energy-based models (US20060034495A1)
Trainable Modular Neural Networks & Document Processing
8 patentsLeCun's mid-career AT&T work produced patents on graph-structured learning systems and colour document image segmentation. The Graph Transformer Networks patent extended backpropagation to modular systems with graph-structured inputs and outputs — a conceptual foundation presaging later graph neural networks.
- Module for constructing trainable modular network — graph-structured data (US6128606A)
- Method and apparatus for extracting foreground and background from colour document image (US5900953A)
Image Processing & Computer Vision
8 patentsThis domain covers general image processing methods including foreground/background extraction, image segmentation, and visual feature analysis. These patents bridge LeCun's foundational character recognition work and his later biometric applications, providing core image manipulation primitives used across multiple product domains.
- Method and apparatus for extracting a foreground image from a color document image (US5900953A)
- Efficient object localisation using Convolutional Networks (NYU research)
Yann LeCun's Highest-Impact IP
The most-cited patents are the foundational Bell Laboratories neural network filings from 1989–1997, with US5900953A leading at 165 citations — reflecting decades of downstream building on LeCun's core methods.
| Patent Number | Title | Year | Citations | Assignee | Status |
|---|---|---|---|---|---|
| US5900953A | Method and apparatus for extracting a foreground image and a background image from a color document image | 1997 | 165 ↑ | AT&T Corp | Inactive |
| US5067164A | Hierarchical constrained automatic learning neural network for character recognition | 1989 | 160 ↑ | AT&T Bell Laboratories | Inactive |
| US20140068740A1 | System and method for biometric authentication in connection with camera equipped devices | 2013 | 145 ↑ | Element, Inc. | Inactive |
| US6128606A | Module for constructing trainable modular network in which each module inputs and outputs data structured as a graph | 1997 | 99 ↑ | AT&T Corporation | Inactive |
| US5105468A | Time delay neural network for printed and cursive handwritten character recognition | 1991 | 91 ↑ | AT&T Bell Laboratories | Inactive |
| US20110218950A1 | Method, system, and computer-accessible medium for classification of at least one ictal state | 2009 | 91 ↑ | New York University | Active |
Yann LeCun's Research Collaborators
Most Frequent Co-Inventors
Collaboration Highlights
The dominance of Element, Inc. collaborators in raw counts reflects the multi-inventor nature of startup IP strategy, where core technology teams are comprehensively listed across all national phase entries of the same underlying invention. Notably, LeCun's earlier Bell Laboratories work was conducted alongside now-legendary researchers including Léon Bottou, Yoshua Bengio, and John S. Denker — representing a remarkable concentration of Turing-calibre researchers on single filings.
- Adam Perold50 joint patents
- Yang Wang46 joint patents
- Sagar Waghmare33 joint patents
- Dushyant Goyal15 joint patents
- Léon Bottou (Bell Labs era)Multiple joint patents
Research Literature by Yann LeCun
349 papers indexed · Output spans foundational CNN architecture, unsupervised learning theory, and broad application domains including vision, language, and multimodal reasoning.
| Title | Year | Citations | Venue / Source |
|---|---|---|---|
| Deep learning | 2015 | 59,186 ↑ | Nature (NYU, Montréal, Toronto, Facebook AI Research, Google) |
| Geometric Deep Learning: Going beyond Euclidean data | 2017 | 2,552 ↑ | EPFL, Facebook AI Research, NYU |
| Efficient object localization using Convolutional Networks | 2015 | 757 ↑ | New York University |
| Efficient Learning of Sparse Representations with an Energy-Based Model | 2007 | 572 ↑ | — |
| Deep learning for AI | 2021 | 553 ↑ | CACM (NYU, Montréal, Toronto) |
Convolutional Network Architecture & Training
Seminal work on CNN design, FFT-based acceleration, feature pooling, and multi-scale processing — the technical core of modern image and speech recognition systems. This cluster includes the foundational papers that established convolutional networks as the dominant paradigm for visual recognition.
Unsupervised & Self-Supervised Learning
Energy-based models, sparse coding, and invariant feature hierarchies — work that directly anticipates contemporary interest in representation learning without labels. This research cluster underpins modern self-supervised pre-training approaches used across language and vision models.
Applications: Vision, Language & Multimodal Reasoning
Spanning pedestrian detection, human pose estimation, video understanding, text classification, and multimodal reasoning — reflecting the breadth of domains to which LeCun's group has applied deep learning methods. Includes the MDETR multimodal detection work (522 citations, 2021).
Patent Jurisdictions
Yann LeCun's patents span 20 jurisdictions, with strong coverage across North America, Europe, Asia-Pacific, and emerging digital identity markets in Southeast Asia, Latin America, and the Middle East.
Filing Markets
The US (21), Canada (19), and EPO (13) represent the core commercial markets for both legacy character recognition technology and recent biometric IP. The strong presence in Southeast Asian markets — Taiwan, Indonesia, Vietnam, Singapore, Malaysia — as well as South Korea, Japan, India, Brazil, Saudi Arabia, and Mexico reflects Element, Inc.'s deliberate strategy to protect mobile biometric authentication technology in high-growth digital identity markets where smartphone penetration is expanding rapidly.
Why Yann LeCun's Portfolio Matters
Strategic implications for patent attorneys, in-house IP teams, and R&D strategists working in AI, computer vision, and identity technology.
FTO Considerations
The Bell Laboratories-era foundational patents — US5067164A, US5105468A, US5058179A, US6128606A — have all lapsed through expiration. The most foundational convolutional neural network architectures for character recognition are now in the public domain and do not represent active FTO obstacles. However, the Element, Inc. biometric authentication and anti-spoofing portfolio presents a live FTO challenge for anyone developing contactless palm-print authentication or 3D liveness detection for mobile devices. With 24 active patents spread across 20 jurisdictions, this is a well-enforced, globally distributed portfolio.
Prior Art Relevance
US5067164A (160 citations, 1989) and US6128606A (99 citations, 1997) are among the most significant prior art references in neural network and machine learning patent prosecution. The 1997 Graph Transformer Networks patent is particularly relevant to modern work on graph neural networks, and examiners in machine learning art units are known to cite the Bell Laboratories character recognition family broadly. Any party filing in deep learning-based image classification, feature extraction from sequential or graph-structured data, or neural network training methods must navigate this prior art landscape carefully.
Yann LeCun Patent Portfolio: Common Questions
How many patents does Yann LeCun hold?
What technology areas does Yann LeCun specialise in?
Which companies hold Yann LeCun's patents?
What are Yann LeCun's most cited patents?
How does Yann LeCun's academic work relate to his patents?
In which countries has Yann LeCun filed patents?
Analyse Yann LeCun's Full Patent Landscape
Access citation maps, FTO analysis, co-inventor networks, and real-time status monitoring across all 159 patents and 20 jurisdictions in PatSnap Eureka IP.
References & External Resources
- United States Patent and Trademark Office — US5067164A: Hierarchical constrained automatic learning neural network for character recognition
- United States Patent and Trademark Office — US5900953A: Method and apparatus for extracting a foreground and background image from a color document image
- European Patent Office — EP2893489B1: System and method for biometric authentication in connection with camera-equipped devices
- WIPO PatentScope — WIPO PatentScope: International patent filings by Yann LeCun
- PatSnap Eureka IP — Full patent portfolio analysis for Yann LeCun
- PatSnap — PatSnap: AI-native global innovation intelligence platform
PatSnap Eureka searches 208M+ patents and papers to answer instantly.