Federated Learning Patents: Leaders, Trends & White Space 2026
- Filing is still accelerating. the 2022 midpoint of 219 filings sat well below the 2025 peak of 739, and growth has not plateaued.
- China and India dominate the filing offices. 930 and 641 filings respectively, dwarfing the United States at 221 and the WIPO/PCT route at 102.
- Momentum is fragmented, not consolidated. the fastest-moving filer this year grew +233% YoY while several established names posted 0% or a full -100% drop-off.
Filing growth compares 2021 (80 records) with 2024 (351) — a three-year span. 2024 is the most recent year we treat as complete: publication lags filing by roughly 18 months, so 2025 onwards are still filling in and any growth rate that ends there would understate the field. Top-5 share is the combined record count of the five largest assignees divided by all 2,019 records in scope (CR5), not by the ranked leaders only.
What this landscape covers
This dataset tracks patent families at the intersection of federated learning, privacy-preserving machine learning and distributed training, filtered to documents whose title, abstract or claims explicitly discuss secure aggregation, differential privacy, client drift, communication overhead or model poisoning, and classified under the AI (G06N3), cryptographic (H04L9) or security (G06F21) IPC groups. That combination isolates the technical core of the field rather than every AI patent that happens to mention federation in passing.
Coverage runs from 2015 through the mid-2026 data cut-off, spanning 2,019 published records. Because publication typically lags filing by around 18 months, the 2026 count of 293 and even the 2025 peak of 739 are undercounts of the true filing activity in those years — the visible trend line will keep rising as later filings publish.
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
Filing trends and technology composition
Two views of the same corpus: how filing volume has moved year over year, and which IPC subclasses carry the claim density.
Filing trend: still climbing
Annual filings rose from zero recorded activity in 2017 to a peak of 739 in 2025, with the 2022 midpoint of 219 confirming the growth curve is convex rather than flattening. The partial 2026 figure of 293 should be read as a floor, not a ceiling, given the usual 18-month publication lag.
Technology composition: AI core, security overlay
G06N (AI model computing) covers 1,838 of the 2,019 records, with G06F (general digital data processing) and H04L (digital information transmission) forming the next layers — the security and networking classes that secure aggregation and communication-efficient training actually depend on. G16H (healthcare informatics) and G06Q (business/admin processing) sitting in the low hundreds mark the two application verticals with the clearest patent presence outside core infrastructure.
Shares are the percentage of the 2,019 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Federated and Privacy-Preserving Learning with Eureka
This page is one run against one query. Ask Eureka your own question about federated and privacy-preserving learning and every answer comes back with the patent numbers behind it.
Try EurekaThe most-cited records and a representative filing
EP4149134A1 — Differential privacy via federated learning with a hierarchical aggregation structure
A method and system for differential privacy using federated learning, comprising: at the lowest level, federated clients holding private data to be provided with differential privacy, located in different zones; at the highest level, a central aggregator learning a model with the federated clients; and at least one intermediate level with a super-node located in each zone, configured to process requests for a local FL model update from a subset of the clients in that zone.Filed by Telefonica Innovacion Digital, S.L.U. — published 2023-03-15.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20210073639A1 | Federated Learning with Adaptive Optimization | 137 |
| 2 | US20210150269A1 | Anonymizing data for preserving privacy during use for federated machine learning | 135 |
| 3 | US20200358599A1 | Private and federated learning | 130 |
| 4 | CN111611610A | 联邦学习信息处理方法、系统、存储介质、程序、终端 | 125 |
| 5 | US20210073677A1 | Privacy preserving collaborative learning with domain adaptation | 123 |
| 6 | CN113434873A | 一种基于同态加密的联邦学习隐私保护方法 | 113 |
| 7 | US20210143987A1 | Privacy-preserving federated learning | 102 |
| 8 | US20200285980A1 | System for secure federated learning | 102 |
| 9 | WO2020229684A1 | Concepts for federated learning, client classification and training data similarity measurement | 95 |
| 10 | US20230308465A1 | System and method for DNN-based cyber-security using federated learning-based generative adversarial network | 91 |
Citation counts inside a searched corpus skew toward older filings simply because they have had more time to accumulate citations — read this as a signal of influence on subsequent filers, not of current commercial importance.
Patent titles are shown in the language they were filed in, not translated, so that each record stays verifiable against the original filing — a translated title will not match in Eureka or in any national register. Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the numbers say about the field
Three patterns cut across the filing trend, the receiving-office split and the citation leaders.
Growth is convex, not linear
The jump from a 2022 midpoint of 219 filings to a 2025 peak of 739 shows the field is still in its acceleration phase rather than approaching saturation. A technology this early in its filing curve has more open claim territory than the raw record count of 2,019 suggests.
Filing activity has shifted east and south
China (930) and India (641) together account for the large majority of receiving-office activity, well ahead of the United States (221), WIPO/PCT (102) and Europe (83). Anyone benchmarking freedom-to-operate needs to weight Chinese and Indian prior art as heavily as US art, not as a secondary check.
Influence sits with a handful of early US filings
The most-cited records — adaptive optimization, anonymization for federated learning, private/federated learning combinations — are clustered around US filings from the 2019-2020 window. That reflects a citation-age effect as much as technical primacy: newer, equally significant filings have not yet had time to accumulate citations.
Co-filing is limited and mostly cross-sector
Only ten co-assignee pairs appear in the dataset, and the strongest of them link a corporate research lab to a university partner rather than two direct competitors. Joint filing is not yet a common strategy in this field, which leaves single-assignee filings as the dominant, and more contestable, pattern.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to federated and privacy-preserving learning, with the prior art for and against each one.
Who is filing, and how momentum is shifting
The assignee table shows a familiar shape for an early-stage field: a mix of large corporates, universities and a few fast-moving newcomers, with momentum concentrated in a small number of accounts rather than spread evenly.
Newer filers are growing fastest, off a small base
The steepest year-over-year growth in the dataset comes from a filer with 20 records in the latest year, up 233% — a reminder that momentum leaders in an early-stage field are often small, not the incumbents with the largest total portfolios.
Established filers have slowed or stalled
Several of the more recognizable names in the corpus show flat or negative year-over-year filing counts in the latest year, including one large filer down 50%. That does not mean these firms have exited the space; it more likely reflects the publication lag pulling their most recent filings out of view.
Corporate-academic pairs lead joint filing
The strongest co-assignee relationship in the dataset links a corporate research arm to a university partner, at four joint filings — still a small number in absolute terms, but the clearest sign of formal collaboration in a corpus otherwise dominated by single-assignee filings.
| Assignee | Recent year | YoY |
|---|---|---|
| SR UNIVERSITY | 20 | +233% |
| VELLORE INSITUTE OF TECH | 12 | 0% |
| Google LLC | 2 | 0% |
| Samsung Electronics Co., Ltd. (Korea) | 1 | -50% |
| VAIBHAV LAXMAN DHASAL (DIRECTOR) | 0 | -100% |
| Qualcomm Incorporated | 0 | -100% |
| International Business Machines Corporation (IBM) | 0 | — |
| Beijing Institute of Technology | 0 | -100% |
Where to take this analysis
The landscape data points to a few concrete next steps depending on what you are trying to decide.
Check freedom-to-operate against the Chinese and Indian filings first
With 930 China filings and 641 India filings against 221 in the US, a freedom-to-operate search that only covers US and EPO art will miss the majority of the relevant prior art in this field.
Explore the assignee table →Treat the 2025-2026 trend line as a floor
The 18-month publication lag means the true 2025 and 2026 filing volumes are higher than the 739 and 293 currently visible. Re-check the trend in six to twelve months before concluding the field has peaked.
Revisit the filing trend →Draft around the under-claimed branches, not the crowded core
Client drift correction, communication-efficient compression and aggregator-side poisoning detection show thinner claim density than the core secure-aggregation and differential-privacy classes, making them better first-filing targets.
Review the white space chips →Common questions about federated learning patents
This dataset contains 2,019 published patent families filed between 2015 and mid-2026 that combine federated learning, privacy-preserving machine learning or distributed training with a specific technical element such as secure aggregation, differential privacy, client drift or model poisoning. The true total is higher than 2,019 once you account for the roughly 18-month lag between filing and publication, which means recent years are undercounted. The figure also reflects a deliberately narrowed search string rather than every patent that mentions federated learning in passing, so broader searches will return larger totals.
China leads with 930 filings in this corpus, followed by India at 641, then the United States at 221, the WIPO/PCT route at 102 and Europe at 83. This is a marked shift from the pattern seen in most earlier-generation AI patent landscapes, where the US and Europe typically led. Anyone assessing competitive risk or freedom-to-operate in this field needs to search Chinese and Indian patent literature as a primary step, not a supplementary one.
It is still accelerating. Annual filings rose from a 2022 midpoint of 219 to a 2025 peak of 739, and the shape of that curve is convex rather than flattening, meaning growth is speeding up rather than slowing down. The partial 2026 figure of 293 looks like a slowdown only because publication lag has not yet caught up with actual filing activity for that year.
The dataset shows a fragmented leadership picture rather than a single dominant filer: some accounts are growing quickly off a small base — one filer grew 233% year over year to 20 filings in the latest year — while several larger, more established names show flat or declining recent-year counts. Total filing volume and recent momentum tell different stories, so a name with a large historical portfolio is not necessarily the one setting the current pace. Co-filing is uncommon across the corpus, with only ten identified co-assignee pairs, suggesting most players are pursuing independent filing strategies.
EP4149134A1, filed by Telefonica Innovacion Digital, describes a hierarchical federated learning architecture that adds an intermediate super-node layer between individual clients and a central aggregator, specifically to deliver differential privacy guarantees at each zone. It matters because it formalizes a three-tier aggregation structure — client, zonal super-node, central aggregator — rather than the simpler two-tier client-to-server model common in earlier federated learning patents. Anyone designing a regionally distributed or telecom-operator-style federated system should check this filing's claim scope before committing to a similar hierarchical aggregation design.
Research Federated and Privacy-Preserving Learning in depth with Eureka
Go past this page: query the whole federated and privacy-preserving learning corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.