Critical Technology

Machine Learning & Foundation Models

The capacity to train, tune and field machine learning at the frontier, foundation and general-purpose models, and the model families across language, vision and science built on them. One of the capabilities of Artificial Intelligence.

International families: 108k2015 to 2024
Leading country: United States1.6× the next
All families: 604khome filings included
Over five years: Rising2020 to 2024 against the five before

Year by year

Patent families by country of origin and the year of the earliest filing, 2000 to 2025; 2025 is a part year, still filling in. International families are those for which protection was sought beyond one country.

Volume

International patent families in the capability, by the year of the earliest filing.

Source: Global Institute for National Capability (GINC), 2026. Underlying patents from 107 patent offices, IFI CLAIMS and Google

Share by country

Each country's share of the year's international families, in per cent: the capability's five largest countries of origin, 2015 to 2024, and the rest of the world.

Source: Global Institute for National Capability (GINC), 2026. Underlying patents from 107 patent offices, IFI CLAIMS and Google

Country leaderboard

International patent families, 2015 to 2024, by country of origin. Growth compares the latest five full years with the five before.

Ten largest

United States

China

South Korea

Japan

Germany

France

Canada

United Kingdom

Taiwan

Sweden

32,358

20,768

15,108

13,257

5,948

2,012

1,987

1,920

1,855

1,430


1	United States	32,358	30.1%	2.0×	71,577
2	China	20,768	19.3%	3.3×	422,113
3	South Korea	15,108	14.0%	3.0×	54,470
4	Japan	13,257	12.3%	1.5×	18,574
5	Germany	5,948	5.5%	2.7×	8,531
6	France	2,012	1.9%	2.7×	2,651
7	Canada	1,987	1.8%	1.9×	2,906
8	United Kingdom	1,920	1.8%	1.7×	2,566
9	Taiwan	1,855	1.7%	2.5×	3,105
10	Sweden	1,430	1.3%	3.7×	1,572
11	India	1,155	1.1%	2.7×	2,238
12	Netherlands	1,148	1.1%	1.4×	1,448
13	Israel	1,040	1.0%	2.2×	1,646
14	Singapore	1,029	1.0%	4.4×	1,261
15	Finland	1,027	1.0%	3.0×	1,128
16	Switzerland	949	0.9%	2.4×	1,178
17	Ireland	576	0.5%	0.9×	1,167
18	Australia	527	0.5%	2.3×	700
19	Cayman Islands	493	0.5%	0.9×	580
20	Türkiye	329	0.3%	9.3×	763
21	Italy	280	0.3%	2.6×	375
22	Russia	226	0.2%	0.8×	379
23	Hong Kong	216	0.2%	1.9×	305
24	Saudi Arabia	210	0.2%	2.9×	340
25	Spain	206	0.2%	3.2×	254
26	Belgium	184	0.2%	1.2×	194
27	Denmark	176	0.2%	2.8×	195
28	Austria	134	0.1%	3.6×	149
29	New Zealand	97	0.1%	4.7×	112
30	Portugal	94	0.1%	7.5×	112
31	Norway	68	0.1%	2.2×	82
32	Brazil	67	0.1%	2.2×	151
33	Malaysia	61	0.1%	2.0×	95
34	Luxembourg	48	0.0%	1.3×	67
35	Poland	39	0.0%	3.9×	88
36	Hungary	24	0.0%	2.0×	32
37	South Africa	18	0.0%	1.3×	22
38	Czechia	17	0.0%	1.1×	44
39	Mexico	12	0.0%	2.0×	35
40	Liechtenstein	10	0.0%	–	10

Applicants

The largest applicants by international patent families, 2015 to 2024: across the world, then within each of the five countries with the most families in Machine Learning & Foundation Models. Names are as held in the source and are not consolidated across subsidiaries.

Global

107,657 international families

SAMSUNG ELECTRONICS CO LTD · South Korea

HUAWEI TECH CO LTD · China

MICROSOFT TECHNOLOGY LICENSING LLC · United States

GOOGLE LLC · United States

BOSCH GMBH ROBERT · Germany

FUJITSU LTD · Japan

4,627

2,552

2,223

1,926

1,913

1,440

United States

32,358 international families

MICROSOFT TECHNOLOGY LICENSING LLC

GOOGLE LLC

IBM

QUALCOMM INC

INTEL CORP

GOOGLE INC

2,207

1,904

1,336

1,181

1,059

978

China

20,768 international families

HUAWEI TECH CO LTD

BEIJING BAIDU NETCOM SCI & TECH CO LTD

PING AN TECH SHENZHEN CO LTD

TENCENT TECH SHENZHEN CO LTD

BEIJING BAIDU NETCOM SCI & TEC

BEIJING BAIDU NETCOM SCIENCE & TECNOLOGY CO LTD

2,404

1,053

929

472

364

South Korea

15,108 international families

SAMSUNG ELECTRONICS CO LTD

LG ELECTRONICS INC

ELECTRONICS & TELECOMMUNICATIONS RES INST

HYUNDAI MOTOR CO LTD

KIA CORP

SEOUL NAT UNIV R&DB FOUNDATION

4,610

1,107

575

529

464

364

Japan

13,257 international families

FUJITSU LTD

NEC CORP

NIPPON TELEGRAPH & TELEPHONE

HITACHI LTD

MITSUBISHI ELECTRIC CORP

TOYOTA MOTOR CORP

1,321

1,220

997

587

574

566

Germany

5,948 international families

BOSCH GMBH ROBERT

SIEMENS AG

ROBERT BOSCH CO LTD

3.6% shared · 5,717 families in both

Japan

United States

China

Germany

South Korea

15,331

13,144

8,455

5,859

5,850

Crosswalk

How Machine Learning & Foundation Models is matched to the patent classifications. The entries below were chosen for it; each stands for itself and everything beneath it in the scheme, 37 Cooperative entries in all. A patent family counts when any of its codes, in either scheme and at any office, falls under one of them. Codes since retired are placed under the entries that replaced them.

Entry	Title in the scheme	Why it is chosen	CPC codes	IPC codes	Documents
G06N3/02	Neural networks	Neural networks: architectures and learning methods	33	29	2,846,752
G06N20/00	Machine learning	Machine learning	3	3	481,425
G06N7/01	Probabilistic graphical models, e.g. probabilistic networks	Probabilistic graphical models	1	1	80,298
G06N3/06	Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons	Left out

CPC and IPC codes are the codes in use on patent documents, at any office, that fall under the entry. Documents is the larger of the two schemes' counts of documents carrying one of them. An entry with no IPC codes can be told apart only in the Cooperative scheme.

The Cooperative codes in use that count towards Machine Learning & Foundation Models: 37 codes on 3,408,475 documents. The full crosswalk is in the cpc-to-capability.csv download.

Code	Title	Under entry	How placed	Documents
G06N3/08	Learning methods	G06N3/02	In the scheme	499,151
G06N3/045	Combinations of networks	G06N3/02	In the scheme	434,000
G06N20/00	Machine learning	G06N20/00	In the scheme	375,730
G06N3/0464	Convolutional networks [CNN, ConvNet]	G06N3/02	In the scheme	344,175
G06N3/09	Supervised learning	G06N3/02	In the scheme	198,361
G06N3/084	Backpropagation, e.g. using gradient descent	G06N3/02	In the scheme	186,248
G06N3/0455	Auto-encoder networks; Encoder-decoder networks	G06N3/02	In the scheme	121,174
G06N3/0442	Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]	G06N3/02	In the scheme	112,568
G06N3/044	Recurrent networks, e.g. Hopfield networks	G06N3/02	In the scheme	99,485
G06N3/04	Architecture, e.g. interconnection topology	G06N3/02	In the scheme	98,739
G06N3/048	Activation functions	G06N3/02	In the scheme	88,885
G06N7/01	Probabilistic graphical models, e.g. probabilistic networks	G06N7/01	In the scheme	80,298
G06N20/20	Ensemble learning	G06N20/00	In the scheme	62,118
G06N3/0499	Feedforward networks	G06N3/02	In the scheme	59,717
G06N3/0475	Generative networks	G06N3/02	In the scheme	57,233
G06N3/047	Probabilistic or stochastic networks	G06N3/02	In the scheme	54,138
G06N3/092	Reinforcement learning	G06N3/02	In the scheme	50,116
G06N3/088	Non-supervised learning, e.g. competitive learning	G06N3/02	In the scheme	49,703
G06N3/042	Knowledge-based neural networks; Logical representations of neural networks	G06N3/02	In the scheme	46,424
G06N3/082	Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections	G06N3/02	In the scheme	44,882
G06N20/10	Machine learning using kernel methods, e.g. support vector machines [SVM]	G06N20/00	In the scheme	43,577
G06N3/096	Transfer learning	G06N3/02	In the scheme	43,567
G06N3/02	Neural networks	G06N3/02	In the scheme	39,334
G06N3/094	Adversarial learning	G06N3/02	In the scheme	39,155
G06N3/0895	Weakly supervised learning, e.g. semi-supervised or self-supervised learning	G06N3/02	In the scheme	36,087
G06N3/0495	Quantised networks; Sparse networks; Compressed networks	G06N3/02	In the scheme	31,938
G06N3/049	Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs	G06N3/02	In the scheme	30,357
G06N3/0985	Hyperparameter optimisation; Meta-learning; Learning-to-learn	G06N3/02	In the scheme	28,127
G06N3/098	Distributed learning, e.g. federated learning	G06N3/02	In the scheme	23,775
G06N3/086	Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming	G06N3/02	In the scheme	8,554
G06N3/10	Interfaces, programming languages or software development kits, e.g. for simulating neural networks	G06N3/02	In the scheme	6,768
G06N3/091	Active learning	G06N3/02	In the scheme	5,858
G06N3/043	Architecture, e.g. interconnection topology based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]	G06N3/02	In the scheme	3,349
G06N3/105	Shells for specifying net layout	G06N3/02	In the scheme	2,971
G06N3/0418	Architecture, e.g. interconnection topology using chaos or fractal principles	G06N3/02	In the scheme	1,017
G06N3/0463	Neocognitrons	G06N3/02	In the scheme	454
G06N3/0409	Adaptive resonance theory [ART] networks	G06N3/02	In the scheme	442

Match: Direct

The capability is itself a technology or a kind of physical system, and the classification names it. Neural-network hardware (G06N3/06) is counted under High-Performance & AI Computing; foundation models have no entry of their own. The crosswalk is a first pass that nobody has reviewed.

Machine Learning & Foundation Models

Year by year

Volume

Share by country

Country leaderboard

Ten largest

Applicants

Global

United States

China

South Korea

Japan

Germany

Similar capabilities

Crosswalk

Other capabilities in Artificial Intelligence