LoRA Patents: Who Leads, Where the Gaps Are 2026
Filing growth compares 2021 (2,162 records) with 2024 (2,846) — 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 41,982 records in scope (CR5), not by the ranked leaders only.
What the Low-Rank Adaptation patent record actually shows
Low-Rank Adaptation, or LoRA, entered patent filings as a fine-tuning technique for large neural models and has since spread into wireless systems, business-process automation and image processing claims that reuse the same low-rank decomposition idea. The corpus behind this page spans 41,982 published records filed or published between 2015 and the 2026 data cut-off, drawn from a search string built around the exact terms low-rank adaptation, LoRA paired with model or neural terms, and low-rank fine-tuning language more generally.
Because publication lags filing by roughly 18 months, the most recent one to two years in any trend understate real filing activity; treat 2024 as the latest year that can be read as complete and everything after it as still filling in.
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
Filing trend and technology composition
Two views of the same 41,982-record corpus: how filing volume has moved year over year, and which IPC subclasses carry the claim density.
Filing trend, 2017–2026
Filings rose from 908 in 2017 to a peak of 2,846 in 2024, a +32% increase over the 2021-2024 span; the 2025 and 2026 figures are still incomplete due to publication lag and should not be read as a slowdown.
Technology composition by IPC subclass
H04W (wireless communication networks) leads at 12.7% of the 41,982 records, followed by H04L (digital information transmission) at 11.6% and G06F (electric digital data processing) at 8.7%; G06N, the subclass most associated with AI model computation, sits at 6.5%. Records can carry multiple classes, so these shares sum past 100%.
Shares are the percentage of the 41,982 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Foundation Models: Low-Rank Adaptation Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about foundation models: low-rank adaptation patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative and most-cited filings
Bi-directional low-rank adaptation for machine unlearning and information retention
The filing describes applying a low-rank adaptation unlearning component to the layers of a machine learning model responsible for specific knowledge, while applying a separate low-rank adaptation retention component to the remaining layers, so a model can selectively forget content without retraining from scratch.Filed by Cisco Technology, published 2026-03-05 as US20260065130A1.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20030229900A1 | Method and apparatus for browsing using multiple coordinated device sets | 3,475 |
| 2 | US20040031058A1 | Method and apparatus for browsing using alternative linkbases | 1,955 |
| 3 | US20120290950A1 | Social-topical adaptive networking (STAN) system allowing for group based contextual transaction offers and a… | 1,488 |
| 4 | US20170173262A1 | Medical systems, devices and methods | 1,227 |
| 5 | US20190349426A1 | The internet of things | 1,108 |
| 6 | US20160045841A1 | New and improved system for processing various chemicals and materials | 861 |
| 7 | US20210144517A1 | Multi-entity resource, security, and service management in edge computing deployments | 845 |
| 8 | US20180165554A1 | Semisupervised autoencoder for sentiment analysis | 797 |
| 9 | US20190209022A1 | Wearable electronic device and system for tracking location and identifying changes in salient indicators of … | 626 |
| 10 | US20090320073A1 | Method and Apparatus for Browsing Using Multiple Coordinated Device Sets | 546 |
Citation counts favour older filings that have had more time to accumulate citations inside the searched corpus; treat them as a signal of influence rather than of current technical importance.
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 mean for filing strategy
Three patterns stand out once the ranking, the trend and the class breakdown are read together.
The top of the field is thin, not dominant
The five leading assignees combined hold 16.7% of all records in scope, and the top ten only reach 20.2%. That is a moderate lead, not a lockout: most of the corpus sits with entities outside the ranked leaders, meaning a new entrant is not filing into a field owned by a handful of players.
Growth is real but recent-year figures are misleading
Filing volume grew from 2,162 records in 2021 to 2,846 in 2024. Every leading assignee tracked for recent-year momentum shows a steep year-on-year drop, but that drop is an artefact of publication lag on 2025-2026 filings, not a genuine pullback from the technique.
LoRA claims sit in infrastructure, not just model architecture
H04W and H04L, both communications subclasses, outrank G06N, the subclass most associated with AI model computation. This suggests a large share of filings apply low-rank adaptation to network or signal-processing contexts rather than to language or vision model fine-tuning specifically.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to foundation models: low-rank adaptation patent landscape, with the prior art for and against each one.
Where to take this next
The landscape points to three concrete follow-ups for a team deciding where to file or diligence.
Check freedom-to-operate against the leader's cluster
With one assignee holding 2,826 records against a fifth-place figure of 368, any filing that touches the leader's core claim area needs a focused FTO check before drafting.
Run an FTO search in EurekaWatch the 2025-2026 filings as they resolve
Because publication lag understates the last 18 months, re-check the trend once 2025 filings finish publishing rather than treating the apparent drop as final.
Set a monitoring alert in EurekaExplore the under-claimed branches directly
Sub-areas like adapter merging and rank-selection automation show thinner filing density than the core fine-tuning claims, which is where new claim scope is easiest to establish.
Explore white space in EurekaCommon questions about the Low-Rank Adaptation patent landscape
It is moderately concentrated but far from locked up. The five leading assignees combined hold 16.7% of the 41,982 records in scope, and the top ten reach only 20.2%. That leaves roughly four-fifths of the corpus spread across the remaining ranked companies and a long tail of smaller filers, so the field is still open to new entrants who avoid the leader's specific claim clusters.
Filing volume grew from 2,162 records in 2021 to a peak of 2,846 in 2024, a 32% increase over that span. The 2025 and 2026 figures appear lower, but that reflects the roughly 18-month lag between filing and publication rather than an actual slowdown; 2024 is the most recent year that can be read as a complete picture.
Wireless communication networks (H04W) and digital information transmission (H04L) carry the largest shares, at 12.7% and 11.6% of the 41,982 records respectively, ahead of general data processing (G06F) at 8.7% and AI-specific computation (G06N) at 6.5%. This means a meaningful share of filings apply low-rank adaptation to network and signal contexts rather than to language or vision model fine-tuning alone, which is worth checking before assuming a filing sits purely in the AI-model space.
The ranking covers 100 companies, with the leading assignee holding 2,826 records against 368 at fifth place and 259 at tenth, showing a steep drop-off after the top filer. Every leading assignee tracked for recent-year momentum also shows a large year-on-year decline, but this is consistent with publication lag rather than an actual pullback in filing activity. Reviewing the leader's claim clusters specifically is more useful than treating the ranked group as a uniform block.
Filing density is comparatively thin in branches such as selective-layer unlearning using low-rank components, multi-tenant adapter merging and isolation, and automated rank-selection for adapter compression, alongside applications of low-rank fine-tuning to business-process systems under G06Q. These sit adjacent to the dense core fine-tuning claims but have not attracted the same volume of filing, making them a reasonable place to look for open claim scope.
Research Foundation Models: Low-Rank Adaptation Patent Landscape in depth with Eureka
Go past this page: query the whole foundation models: low-rank adaptation patent landscape 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.