Book a demo

Cut patent&paper research from weeks to hours with PatSnap Eureka AI!

Try now

Demis Hassabis Patents & Innovation Profile — PatSnap Eureka

Demis Hassabis Patents & Innovation Profile — PatSnap Eureka
Inventor Profile · PatSnap Eureka

Demis Hassabis: Patent Portfolio & Innovation Analysis

Demis Hassabis is the co-founder and CEO of Google DeepMind, holding 23 patents assigned exclusively to DeepMind Technologies Limited, spanning neural network architectures, visual concept learning, and AI-driven algorithm optimisation for hardware devices, with filings across 8 jurisdictions from 2018 to 2025. His published research record of 122 papers — including the AlphaFold family with over 10,500 citations — constitutes a critical prior art corpus for any IP professional operating in the machine learning, computer vision, or AI-for-science sectors.

23
Patents
2018–2025
Years Active
8
Jurisdictions

Patent Filing Activity

Peak year was 2018 with 11 filings, corresponding to the visual concept learning family filed across US, EP, WO, AT, and DE jurisdictions.

Annual Patent Filings by Demis Hassabis: 2018=11, 2020=1, 2022=3, 2023=7, 2025=1 Line chart showing Demis Hassabis's patent filing activity by year, derived from PatSnap Eureka patent database. Peak year was 2018 with 11 filings. 11 8 6 3 0 2018 2020 2022 2023 2025 11 1 3 7 1
📋
23
Total Patents
14 active · 5 pending
📅
2018–2025
Filing Period
Active engagement with IP system
🌐
8
Jurisdictions
US, EP, TW, WO, DE, AT, CN, IN
🏢
DeepMind
Primary Assignee
23 patents assigned
🔬
Neural Networks
Top Technology
10 patents in G06N3 domain
Patent Analytics

Demis Hassabis's Patent Filing Patterns

A two-phase filing strategy: an 11-patent burst in 2018 for visual concept learning, followed by a 2022–2023 acceleration for algorithm optimisation, mirroring DeepMind's published research milestones.

Annual Patent Filings

Peak year 2018 saw 11 filings across 5 jurisdictions from a single priority date; 2023 saw 7 filings in the algorithm optimisation family.

Annual Patent Filings by Demis Hassabis: 2018=11, 2020=1, 2022=3, 2023=7, 2025=1 Line chart showing Demis Hassabis's patent filing activity by year, derived from PatSnap Eureka patent database. Peak year was 2018 with 11 filings. 11 8 6 3 0 2018 2020 2022 2023 2025 11 1 3 7 1

Technology Domain Breakdown

Neural Networks & Deep Learning (G06N3) accounts for 50% of the portfolio, with Computer Vision and Knowledge Representation as secondary clusters.

Technology Domain Breakdown for Demis Hassabis: G06N3 Neural Networks=50%, G06N5 Knowledge Representation=20%, G06V10 Computer Vision=15%, G06V20 Computer Vision Scene=10%, G06F30 Computer-Aided Design=5% Donut chart showing the distribution of Demis Hassabis's patents across technology domains based on IPC classification codes from PatSnap Eureka. 23 patents Neural Networks (50%) Knowledge Rep. (20%) Computer Vision (15%) CV Scene (10%) CAD (5%)

Search Demis Hassabis's full patent landscape in PatSnap Eureka IP

Explore in PatSnap Eureka IP →
Technology Domains

Core Areas of Innovation

Demis Hassabis's patent portfolio concentrates on two architecturally significant clusters: neural network-based visual concept learning and AI-driven algorithm optimisation for hardware devices.

Neural Networks & Deep Learning

10 patents

Covers the architecture and training of artificial neural networks with a focus on disentangled visual concept learning — enabling systems to map symbolic language inputs to generated visual outputs by decomposing scenes into disentangled latent representations. Also includes abstraction learning via hidden layer activation patterns.

  • Learning visual concepts using neural networks (US20200234468A1)
  • Learning visual concepts using neural networks (US11354823B2)
  • Learning abstractions using patterns of activations of a neural network hidden layer (US12333435B1)
IPC: G06N3

Knowledge Representation & Reasoning

4 patents

Covers machine reasoning and knowledge representation systems, specifically AI-guided search over candidate algorithm spaces using learned knowledge captured through tensor decompositions. Addresses the automation of algorithm discovery — replacing manual engineering with reinforcement learning-based search.

  • Optimizing algorithms for hardware devices (WO2024074452A1)
  • Optimizing algorithms for hardware devices (US20240127045A1)
IPC: G06N5

Computer Vision

5 patents

Five patents under G06V10 and G06V20 address the generation and understanding of visual scenes guided by symbolic language inputs, relevant to image synthesis, visual question answering, and multimodal AI systems that interpret and produce visual content from linguistic prompts.

  • Learning visual concepts using neural networks (WO2019011968A1)
  • Learning visual concepts using neural networks (EP4053740A1)
IPC: G06V10, G06V20
Most Cited Patents

Demis Hassabis's Highest-Impact IP

The 2018 "Learning visual concepts using neural networks" family dominates citations, with US20200234468A1 accumulating 19 forward citations — indicating foundational methodology in disentangled visual representation learning.

Patent Number Title Year Citations Assignee Status
US20200234468A1 Learning visual concepts using neural networks 2018 19 ↑ DeepMind Technologies Limited Active
US11354823B2 Learning visual concepts using neural networks 2018 7 ↑ DeepMind Technologies Limited Active
WO2019011968A1 Learning visual concepts using neural networks 2018 6 ↑ DeepMind Technologies Limited
US20240127045A1 Optimizing algorithms for hardware devices 2022 4 ↑ DeepMind Technologies Limited Pending
US12333435B1 Learning abstractions using patterns of activations of a neural network hidden layer 2023 2 ↑ DeepMind Technologies Limited Active
WO2024074452A1 Optimizing algorithms for hardware devices 2023 1 ↑ DeepMind Technologies Limited
View All 23 Patents with Full Citation Analysis
Access the complete citation analysis, full patent text, assignee history, and claim mapping in PatSnap Eureka IP.
EP4053740A1 — Learning visual concepts Full claim text & prosecution history + 16 more patents
Analyse in PatSnap Eureka IP →
Collaboration Network

Demis Hassabis's Research Collaborators

Most Frequent Co-Inventors

Top Co-Inventors of Demis Hassabis: Alexander Lerchner=14 joint patents, Matej Balog=12, Christopher Paul Burgess=12, Irina Higgins=12, Arka Tilak Pal=12, Nicolas Sonnerat=12, Loic Matthey=12, Matthew Botvinick=11 Horizontal bar chart showing the most frequent co-inventors in Demis Hassabis's patent portfolio based on PatSnap Eureka data. Lerchner 14 Balog 12 Burgess 12 Higgins 12 Pal 12 Sonnerat 12 Botvinick 11

Collaboration Highlights

Hassabis's patent collaborations fall into two distinct teams: the visual concept learning family is led by Alexander Lerchner and Christopher Paul Burgess, while the algorithm optimisation family centres on Matej Balog, David Silver, and Julian Schrittwieser from DeepMind's reinforcement learning group. The depth of co-authorship across both families indicates substantive technical involvement rather than supervisory naming.

  1. Alexander Lerchner 14 joint patents
  2. Matej Balog 12+ joint patents
  3. Christopher Paul Burgess 12 joint patents
  4. Irina Higgins 12 joint patents
  5. Matthew Botvinick 11 joint patents
Map the Full Network in PatSnap Eureka IP
Academic Contributions

Research Literature by Demis Hassabis

122 papers indexed · Research clusters across AI for structural biology (AlphaFold), AI-driven algorithm and mathematical discovery (AlphaTensor, AlphaDev), and neuroscience-inspired reinforcement learning.

Title Year Citations Venue / Source
Accurate structure prediction of biomolecular interactions with AlphaFold 3 2024 10,525 ↑ Google DeepMind / Isomorphic Labs
AlphaFold Protein Structure Database 2021 6,044 ↑ DeepMind / EMBL-EBI
Protein complex prediction with AlphaFold-Multimer 2021 2,643 ↑ DeepMind
Improved protein structure prediction using potentials from deep learning 2020 2,481 ↑ DeepMind / Francis Crick Institute / UCL
AlphaFold Protein Structure Database in 2024 2023 1,599 ↑ Google DeepMind / Seoul National University / EMBL

AI for Structural Biology

The AlphaFold family (2020–2024) represents the most scientifically impactful body of work, with AlphaFold 3 alone accumulating over 10,500 citations since May 2024. These papers have reshaped drug discovery, protein engineering, and computational biology at a global scale.

AI-Driven Algorithm Discovery

The AlphaTensor (2022, 370 citations) and AlphaDev (2023, 127 citations) papers directly underpin the algorithm optimisation patents filed from 2022 onwards — one of the clearest examples of an inventor's literature directly foreshadowing their patent strategy.

Neuroscience-Inspired Reinforcement Learning

Earlier work on hippocampal memory, spatial cognition, and episodic memory forms the conceptual foundation for DeepMind's reinforcement learning architectures and memory-augmented neural network designs, including the "Reinforcement Learning, Fast and Slow" paper (2019, 423 citations).

Global Footprint

Patent Jurisdictions

DeepMind's filing strategy for the Hassabis-named portfolio covers 8 jurisdictions, with the US and EP as primary markets and Taiwan specifically targeted for the hardware algorithm optimisation family.

Patent Jurisdictions for Demis Hassabis: US=7, EP=5, TW=3, WO=2, DE=2, AT=2, CN=1, IN=1 Horizontal bar chart showing the distribution of Demis Hassabis's patents by country/jurisdiction based on PatSnap Eureka data. United States 7 Europe (EP) 5 Taiwan 3 PCT / WO 2 Germany 2 Austria 2 China 1 India 1

Filing Markets

The US (7 patents) and EP (5 patents) serve as the primary commercial markets, while Taiwan's 3 patents are particularly telling — algorithm optimisation for hardware devices has been specifically filed in a jurisdiction dominated by TSMC, MediaTek, and other chipmakers. India's 2025 pending application signals expanding geographic ambition for the algorithm optimisation family.

🇺🇸 United States · 7 🇪🇺 Europe (EP) · 5 🇹🇼 Taiwan · 3 🌐 PCT / WO · 2 🇩🇪 Germany · 2 🇦🇹 Austria · 2 🇨🇳 China · 1 🇮🇳 India · 1
For IP Professionals

Why Demis Hassabis's Portfolio Matters

Strategic implications for patent attorneys, in-house IP teams, and R&D strategists working in machine learning, computer vision, and AI for science.

⚠️

FTO Considerations

The visual concept learning family (US11354823B2, US20200234468A1, EP4053740A1 and related grants) covers methods for training neural networks to disentangle visual concepts and generate images from symbolic inputs. Organisations developing multimodal AI systems, text-to-image models, or symbol-grounded visual reasoning systems should assess whether their architectures engage the claim scope of these active patents. The 2023 continuation US12333435B1 extends this family's effective life to at least 2043.

🔍

Prior Art Relevance

Hassabis's 122-paper research record constitutes an exceptional body of prior art. The most cited patent US20200234468A1 has accumulated 19 forward citations. Papers predating his patent applications — particularly neuroscience and early deep learning work from his UCL years — may be relevant to novelty and obviousness assessments against third-party patents in overlapping domains. The AlphaFold literature (10,525+ citations) is a critical reference for structural biology and drug discovery IP searches.

Unlock Licensing & Monitoring Insights
Access full FTO analysis, licensing opportunity mapping, and portfolio monitoring in PatSnap Eureka IP.
Licensing opportunities Monitoring alerts Full FTO report
Access in PatSnap Eureka IP →
Frequently Asked Questions

Demis Hassabis Patent Portfolio: Common Questions

Demis Hassabis holds 23 patents across 8 jurisdictions (US, EP, TW, WO, DE, AT, CN, IN), with filings from 2018 to 2025. The primary assignee is DeepMind Technologies Limited. The portfolio includes 14 active patents, 5 pending applications, and 2 inactive patents.
His patents concentrate on two primary areas: neural network-based visual concept learning (methods for training networks to disentangle and generate visual concepts from symbolic inputs, G06N3 — 10 patents) and the use of AI to optimise algorithms for specific hardware devices through tensor decomposition and reinforcement learning-based search (G06N5 — 4 patents). Computer vision (G06V10, G06V20) accounts for a further 5 patents.
All 23 patents in this dataset are assigned to DeepMind Technologies Limited, the London-based AI research company he co-founded in 2010 and which was acquired by Google in 2014. Some US records also list GDM Holding LLC as a co-assignee, reflecting Google's corporate structure.
His most frequent co-inventors include Alexander Lerchner (14 joint patents), Matej Balog (12+ joint patents), Christopher Paul Burgess, Irina Higgins, Arka Tilak Pal, Nicolas Sonnerat, and Loic Matthey-de-l'Endroit (each approximately 12), and Matthew Botvinick (11). These collaborators correspond to the two core patent families in his portfolio.
The most cited patent is US20200234468A1, "Learning visual concepts using neural networks" (2018), with 19 forward citations. The same underlying invention protected via US11354823B2 (7 citations) and WO2019011968A1 (6 citations) adds a further 13 citations across those records. All are assigned to DeepMind Technologies Limited.
There is a direct but selective relationship. The algorithm optimisation patents (2022–2023) correspond closely in timing and content to the published AlphaTensor and AlphaDev papers. The visual concept learning patents align with DeepMind's disentangled representation research. However, Hassabis's most-cited research — the AlphaFold protein structure prediction work, which has accumulated over 10,500 citations — does not appear in his personal named-inventor patent record, suggesting that AlphaFold's IP is managed separately or primarily through open publication.

Analyse Demis Hassabis's Full Patent Landscape

Search 208M+ patents and papers. Map the complete DeepMind portfolio, track prosecution status, and conduct FTO analysis — all in PatSnap Eureka IP.