Fastest
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 →https://www.patsnap.com/resources/blog/rd-blog/cross-modal-retrieval-and-alignment-patent-landscape/ · Patsnap · data cut-off 2026-07-31 · downloaded from the live page
Patent Landscape · Multimodal ModelsCross Modal Retrieval and Alignment Patents: Filing Trends and Leading Assignees
- Concentrated at the top. Five assignees hold 57.8% of all 64 records in scope, with a long tail of single-filing entrants below tenth place.
- Filing kept climbing through the last complete year. annual filings rose from 6 in 2021 to 10 in 2024, a 67% increase, before the count of published records dips in 2025-2026 purely from publication lag.
- Claim space skews toward general-purpose infrastructure. G06F and G06N subclasses each cover more than a third of records, while speech (G10L) and bioinformatics (G16B) sit at just 3.1% each.
64
Published Records
58%
Top-5 Share of All Records
+67%
Filing Growth 2021→2024
US
Leading Jurisdiction
Filing growth compares 2021 (6 records) with 2024 (10) — 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 64 records in scope (CR5), not by the ranked leaders only.
Published byPatsnap Research··7 min readSourced from Patsnap Eureka
Overview
What this landscape covers
Cross modal retrieval and alignment patents cover methods for mapping images, text, audio and other modalities into a shared embedding space so that a query in one modality can retrieve matches in another. The search string used to build this dataset combines retrieval and alignment concepts with training mechanics such as contrastive pretraining, negative sampling and zero shot transfer, isolating filings that describe the underlying matching mechanism rather than a specific downstream application.
The 64 records in scope span filings from 2015 through the 2026 cut-off, filed across the United States, the WIPO PCT route, Europe, India, Australia and Canada. Because publication trails filing by roughly 18 months, the most recent one to two years in the trend understate actual filing activity and should be read as provisional.
Filing activity and technology composition, 2015-2026
Source: Patsnap Eureka. Assignee ranking and totals. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Data
Filing trends and technology composition
The record set spans 2015 through the 2026 cut-off and is classified across eight IPC subclasses, reflecting a field still anchored in general-purpose computing and AI infrastructure rather than a single applied vertical.
A three-year run of real growth, then a publication-lag dip
Filings rose from 6 in 2021 to 10 in 2024, a 67% increase, with 2025 the peak observed year at 20 records. The apparent fall in 2026 reflects the roughly 18-month gap between filing and publication, not a slowdown in filing activity.
Infrastructure classes dominate over applied verticals
G06F (electric digital data processing) and G06N (AI-based computing) each appear in over a third of the 64 records, and G06V (image/video recognition) in nearly 30%. Narrower application classes such as speech and audio (G10L) and bioinformatics (G16B) each cover only 3.1% of records, marking them as comparatively open ground.
Shares are the percentage of the 64 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Source: Patsnap Eureka. Filing trend and technology composition. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.
Go deeper on Cross Modal Retrieval and Alignment with Eureka
This page is one run against one query. Ask Eureka your own question about cross modal retrieval and alignment and every answer comes back with the patent numbers behind it.
Try EurekaKey Patents
Most-cited records in the corpus
Representative filing
Efficient zero-shot event extraction with context-definition alignment
TENCENT AMERICA LLC
A zero-shot event extraction model training method obtains a raw text corpus and verbal synsets, extracts event type definitions from the synsets, and generates alignment data by pairing corpus sentences with those definitions. A first encoder is trained on that alignment data to embed target mentions into a shared embedding space, and a second encoder is trained to embed the event type definitions into the same space, producing a zero-shot extraction model that can classify a candidate mention against definitions it never saw during training.Filed by Tencent America LLC, published 2026-01-01.
Highest-citation records in scope
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20200311798A1 | Search engine use of neural network regressor for multi-modal item recommendations based on visual semantic e… | 105 |
| 2 | US20230306761A1 | Methods for identifying cross-modal features from spatially resolved data sets | 30 |
| 3 | WO2022051546A1 | Methods for identifying cross-modal features from spatially resolved data sets | 16 |
| 4 | WO2026024917A1 | Systems, methods, kits, and apparatuses for know your model systems in value chain networks | 12 |
| 5 | WO2026024918A1 | Ai-based energy edge platforms, systems, and methods | 10 |
| 6 | WO2026024858A1 | Configured artificial intelligence systems and methods for software-defined vehicles | 10 |
| 7 | US12469483B1 | Comprehensive AI-enabled systems for immersive voice, companion, and augmented/virtual reality interaction so… | 10 |
| 8 | US20250148218A1 | Automatically generating descriptions of augmented reality effects | 10 |
| 9 | US12131365B2 | Search engine use of neural network regressor for multi-modal item recommendations based on visual semantic e… | 9 |
| 10 | US20250148816A1 | Model fine-tuning for automated augmented reality descriptions | 8 |
Citation counts favour older records simply by virtue of longer exposure time; treat them as a signal of influence within this corpus, not as a current-importance ranking.
Each row carries its publication number; clicking a row searches Eureka by that number.
Source: Patsnap Eureka. Citation counts and representative records. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Run it yourself
Put your own technology through the same analysis
Where to run it
For builders
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 →Insights
What the numbers mean for filing strategy
Three patterns stand out once the record set is broken down by assignee, technology class and receiving office.
Concentration
57.8% held by top 5
of 64 records
A short leader group, then a long tail
The leading assignee holds 12 records; by fifth place that has fallen to 3, and by tenth place to 1. Combined, the top 5 account for 57.8% of all 64 records and the top 10 for 70.3%, leaving 37 companies competing for the remaining share with mostly single-digit filing counts.
37 companies ranked in total
Momentum
6 → 10 filings, 2021-2024
+67% over three years
Growth is real, but 2025-2026 are undercounted
Filings grew 67% from 2021 to 2024, the last year that can be treated as a complete publication cohort. The 2025 figure of 20 records is the highest observed so far, but because publication lags filing by around 18 months, both 2025 and 2026 will keep rising in later data pulls.
2024 is the last complete year
Composition
39.1% in G06F
of 64 records
Filings cluster in general digital processing and AI classes
G06F and G06N together anchor the bulk of activity, each covering more than a third of the 64 records, with G06V close behind at 29.7%. Business-process framing (G06Q) and image-generation framing (G06T) each sit at exactly 25.0%, showing the technology is being claimed as much for commerce workflows as for core vision pipelines.
Classes overlap; shares exceed 100% combined
Eureka AI Agent
Looking for what nobody has claimed yet?
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to cross modal retrieval and alignment, with the prior art for and against each one.
Source: Patsnap Eureka. Co-assignee relationships and derived observations. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Players
Who is filing, and where the gaps sit
Filing activity is led by a small set of technology and healthcare-adjacent assignees, with most of the ranked list holding only one or two records each.
Leader
12 records
top assignee
A single assignee holds the largest share
The leading assignee's 12 records are more than double the fifth-place count of 3, marking a clear gap between the top of the ranking and everyone below it.
Ranked leader of 37 companies
Recent momentum
2 filings, latest year
Snap Inc
Snap is the most active filer in the latest observed year
Snap Inc recorded 2 filings in the latest year, ahead of Google and Strong Force EE Portfolio 2022 LLC, which each show 1 filing and flat year-over-year momentum. Microsoft Technology Licensing shows a -100% year-over-year change, though this may reflect publication lag rather than a genuine stop in filing.
Momentum read against the last observed year
Long tail
37 companies ranked
whole ranking returned
Most assignees hold only one or two records
Below the top 10, which together hold 70.3% of the 64 records, the remaining assignees are spread thinly, each contributing a small number of filings. This is typical of a technology still being explored by many entrants rather than locked down by a handful of incumbents.
70.3% held by top 10 combined
Recent-year filing momentum by assignee
| Assignee | Recent year | YoY |
|---|---|---|
| Snap Inc | 2 | — |
| Google LLC | 1 | 0% |
| STRONG FORCE EE PORTFOLIO 2022 LLC | 1 | 0% |
| The General Hospital Corporation | 0 | — |
| Siemens AG | 0 | — |
| Microsoft Technology Licensing, LLC | 0 | -100% |
| The Board of Trustees of the University of Illinois | 0 | — |
| TRAN BAO | 0 | -100% |
Source: Patsnap Eureka. Assignee-level momentum. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
What's Next
Where to take this analysis
The dataset points to a field still open at the edges even as a handful of assignees build depth at the centre.
Watch the 2025-2026 cohort as it fills in
Because publication lags filing by roughly 18 months, the true size of the 2025 and 2026 cohorts will only become clear in later data pulls. Re-running this search in six to twelve months will sharpen the growth read beyond the confirmed 2021-2024 trend.
Track filing trends in EurekaProbe the under-claimed branches before they fill
Speech, audio and bioinformatics applications of cross-modal alignment each sit under 4% of records despite the core matching techniques being well established. A claim scoped to one of these verticals has more open ground than a general shared-embedding-space claim.
Map white space in EurekaSource: Patsnap Eureka. Forward-looking reading of the same dataset. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
FAQ
Common questions on cross modal retrieval patents
These are filings that describe mapping two or more modalities, such as image and text, into a shared embedding space so that a query in one modality can retrieve matches in another. The dataset behind this page specifically targets filings that combine that retrieval or alignment concept with training mechanics like contrastive pretraining, negative sampling or zero-shot transfer. This excludes filings that merely mention multimodal data in passing without claiming the matching mechanism itself.
The ranked list covers 37 companies, with the leading assignee holding 12 of the 64 records in scope. That lead narrows quickly: by fifth place the count drops to 3 records, and the top 5 combined hold 57.8% of all records. Below tenth place, most of the remaining assignees hold only one or two filings each, which is typical of a technology area still attracting new entrants.
Filing activity grew from 6 records in 2021 to 10 in 2024, a 67% increase over that three-year span, and 2024 is the most recent year that can be treated as a complete publication cohort. The apparent decline in 2025 and 2026 figures is an artefact of publication lag, since patent applications typically take around 18 months to publish after filing. Readers should treat any 2025-2026 count as a floor, not a ceiling.
Based on IPC subclass coverage across the 64 records, speech and audio analysis (G10L) and bioinformatics (G16B) each appear in only 3.1% of records, far below the 39.1% and 37.5% held by general digital processing (G06F) and AI-based computing (G06N). This suggests core retrieval and alignment mechanisms are well covered in general-purpose form, but their application to speech, audio and bioinformatics workflows remains comparatively open for new claims.
Among receiving offices in this dataset, the United States leads with 26 filings, followed by the WIPO PCT route with 15 and the European Patent Office with 9. India, Australia and Canada each show smaller counts, with India at 7 and Australia and Canada at 2 each. The heavy use of the PCT route alongside direct US filing suggests many applicants are pursuing multi-jurisdiction protection rather than filing in a single market.
Answers are grounded in the same dataset. Derived from a Patsnap search on Cross Modal Retrieval and Alignment covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Research Cross Modal Retrieval and Alignment in depth with Eureka
Go past this page: query the whole cross modal retrieval and alignment 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.