Machine Learning Patents from Germany

Inflection point 2017: Filing activity more than doubled from 155 families (2016) to 355 (2017), coinciding with the global deep learning wave triggered by breakthroughs in image recognition and natural language processing. Growth continued to a peak of 1,595 families in 2021. The 2022–2023 plateau at ~1,400–1,600 families likely reflects maturation rather than decline.

Neural Networks dominate with over half of all families (4,427). The Quantum Computing segment (461 families) is still small but grew from 2 families in 2014 to 176 in 2023 — an 88x increase, making it the fastest-growing sub-area by far.

Paradigm shift: In 2014, Knowledge-based systems (42 families) led ahead of Neural Networks (24). By 2017, Neural Networks had surged past all other areas, and by 2023 they accounted for 842 families — 35x growth in under a decade. Pattern Recognition (G06F18) emerged as a distinct category only from 2019, coinciding with the revised IPC 2019.01 edition. Quantum Computing became a measurable force from 2020 onward.

CNNs (138) and Auto-encoders (141) lead the architecture types, while Generative Networks / GANs (94) and Recurrent Networks / RNN (73) follow. Among learning methods, Supervised learning (149) dominates, but Reinforcement Learning (89), Transfer Learning (66), and Federated Learning (57) show strong emerging interest. The Federated Learning segment is particularly noteworthy for privacy-preserving AI applications in automotive and healthcare.

Robert Bosch dominates with 1,835 families — more than the next two combined (Siemens + SAP = 1,479). The automotive sector accounts for 8 of the top 15 filers. SAP is the outlier: it is the only top filer where general ML and Knowledge-based systems outweigh Neural Networks, reflecting its enterprise and business-process focus. ZF Friedrichshafen has the highest Neural Network concentration (83%) of any top filer.

Computer Vision (G06V, 1,365 families) and Image Processing (G06T, 1,103) are the #1 application area — combined, they appear on more ML families than any single sub-area. Vehicle Control (B60W, 746) and Traffic Systems (G08G, 355) confirm the automotive dominance. Healthcare (G16H + A61B = 862) is the second-largest application domain, driven by Siemens Healthcare and Siemens Healthineers. Robotics (B25J, 133) and Speech Recognition (G10L, 164) round out the industrial landscape.

Dramatic geographic shift: In 2014, the US Patent Office was the #1 filing destination (67 apps vs. 17 DE). By 2022, the German Patent Office (DPMA) overtook the USPTO (456 vs. 404), and by 2023, domestic filings led all routes (600 DE vs. 404 WO vs. 313 US). This reflects both a strategic shift toward PCT filings for broader international coverage and a growing domestic-first approach. South Korea (KR) is a consistent secondary market, particularly for automotive-related ML.

IBM is the #1 external R&D partner for German ML applicants with 220 joint families. Huawei (128 families) is the second-largest — a strong China connection reflecting joint R&D operations. International subsidiaries (Siemens US/China, Bosch India) represent global R&D coordination within German multinationals. Academic collaborations are notable: Tsinghua University (45), Carnegie Mellon (24), and Nanyang Tech (21) span three continents of research partnerships.

Bosch is the collaboration king — partnering with Tsinghua (45 families), Uni Freiburg (37), and Carnegie Mellon (24), spanning three continents. The VW Group shows intense internal knowledge-sharing between Audi, Porsche, VW, and their software unit CARIAD. A notable emerging pair: Eleqtron x Uni Siegen (11 families) represents the quantum computing startup ecosystem. The Bosch-Tsinghua axis alone accounts for more co-filings than the entire VW Group's internal collaboration.

The USPTO leads with a 56.1% grant rate — more than double the European (21.6%) and German (22.0%) rates. This reflects both the higher volume of US filings by German applicants and the USPTO's historically faster examination process for software-related patents. The DPMA and EPO rates are comparable (22% vs. 21.6%), suggesting consistent European examination standards. Many recent filings (2022-2024) are still under examination, so these rates will increase over time.

Siemens Healthcare dominates citation impact with 25.1 citations per family — the highest citation density of any large German ML filer. 6 of the top 10 most-cited families belong to Siemens Healthcare/Healthineers, mostly in medical imaging AI. Autonomos GmbH (an autonomous driving startup) holds the #2 most-cited family (399 citations). BASF (13.4 avg) and Fraunhofer (15.4 avg) punch above their portfolio size, suggesting high-quality, foundational patents. In contrast, Bosch's large portfolio (921 cited families) shows a more applied, incremental pattern at 7.6 citations per family.

92% of hidden ML families are in image processing (933 of 1,015) — patents that use neural networks for image analysis but are classified under G06T or G06V rather than G06N. Siemens Healthcare alone accounts for 355 hidden families (234 + 121 Healthineers), reflecting their deep medical imaging AI work. The hidden patent count is growing faster than the core count: from 4 families in 2014 to 303 in 2023, roughly tripling the proportion from 4% to 19% of the combined total. This means the "true" German ML patent landscape is larger than what classification-only searches reveal.