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/model-compression-for-deployment-targets-patent-landscape/ · Patsnap · data cut-off 2026-07-31 · downloaded from the live page
Patent Landscape · 533 RecordsModel Compression for Deployment Targets: Patent Trends and Who Leads
- 24.0% concentration at the top. The top five assignees hold 128 of 533 records, but the leader alone (41 records) shows zero filings in the latest year — recent momentum and cumulative claim strength do not track together.
- Filing has plateaued, not fallen. 2021's 77 records dropped to 61 by 2024, a -21% change, right as the field's peak year (2025, 108 records) is still filling in under the usual 18-month publication lag.
- Audio and transmission branches stay thin. G10L (speech/audio, 4.9%) and H04L (digital transmission, 3.8%) trail far behind the dominant G06N class (73.4%), marking where compression claims tied to codec or link constraints remain open.
533
Published Records
24%
Top-5 Share of All Records
-21%
Filing Growth 2021→2024
US
Leading Jurisdiction
Filing growth compares 2021 (77 records) with 2024 (61) — 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 533 records in scope (CR5), not by the ranked leaders only.
Published byPatsnap Research··7 min readSourced from Patsnap Eureka
Overview
What this landscape covers
This landscape tracks 533 published patent records filed between 2015 and the 2026-07-31 data cut-off, matched against claim language covering model compression, weight clustering and mixed-precision quantization alongside deployment-side constraints such as a target hardware profile, latency budget, accuracy constraint or neural architecture search. It is a field defined less by a single invention than by a combination claim pattern: a compression mechanism paired with an explicit deployment constraint.
Filings run across six receiving offices, led by the United States and the PCT route through WIPO, with Europe, India, China and Canada following. The technology composition leans heavily on AI-model claim classes, which frames most of this field as core machine-learning architecture rather than a narrow hardware-engineering niche.
Filing volume and technology composition, 2015-2026
Source: Patsnap Eureka. Assignee ranking and totals. Derived from a Patsnap search on Model Compression for Deployment Targets 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
Filing Data
Filing trends and technology composition
The dataset covers 533 published records from 2015 through the 2026-07-31 cut-off, spanning national and PCT filings across six receiving offices.
Filing activity climbed then plateaued
Annual filings rose from 2 records in 2017 to a peak of 108 in 2025. The only complete multi-year comparison available runs from 77 records in 2021 to 61 in 2024, a -21% change — read as a plateau in a maturing filing base, not a decline, since 2025-2026 figures are still filling in under the usual 18-month publication lag.
AI-model classes dominate; audio and transmission branches stay thin
G06N (AI computing models) appears on 73.4% of the 533 records and G06F (digital data processing) on 30.6%, confirming that most compression claims are framed as core model or processing architecture. Image/video (G06V, 11.1%), data recognition (G06K, 7.3%), image processing (G06T, 5.4%), speech/audio (G10L, 4.9%), business processing (G06Q, 4.3%) and digital transmission (H04L, 3.8%) trail well behind, marking the thinner technical branches.
Shares are the percentage of the 533 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 Model Compression for Deployment Targets covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.
Go deeper on Model Compression for Deployment Targets with Eureka
This page is one run against one query. Ask Eureka your own question about model compression for deployment targets and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative Filing
A representative record and the most-cited work
Representative Record
WO2024127462A1 — Automatic machine learning development system, automatic machine learning development method and program
NEC CORPORATION
An automatic machine-learning development system built around four blocks: a neural architecture search block that outputs a trained small CNN with a derived backbone and alpha parameter values from a larger CNN model; an alpha-analyzer block that narrows the candidate search space; a model-compression block that replaces sublayers within that narrowed space to produce a lightweight CNN; and a deployment block for the resulting model.Filed by NEC Corporation, 2026-06-20 publication record referenced in this dataset.
Most-cited records in scope
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20220237368A1 | Systems and methods for machine content generation | 269 |
| 2 | US20240022927A1 | Systems, methods, and apparatus on wireless network architecture and air interface | 174 |
| 3 | US20050088435A1 | Novel 3D ear camera for making custom-fit hearing devices for hearing aids instruments and cell phones | 162 |
| 4 | US20230252224A1 | Systems and methods for machine content generation | 156 |
| 5 | US20220036194A1 | Deep neural network optimization system for machine learning model scaling | 155 |
| 6 | US20210271934A1 | Method and System for Predicting Wildfire Hazard and Spread at Multiple Time Scales | 105 |
| 7 | US9715495B1 | Topic-influenced document relationship graphs | 93 |
| 8 | US20110307438A1 | High-dimensional data analysis | 84 |
| 9 | US20220036123A1 | Machine learning model scaling system with energy efficient network data transfer for power aware hardware | 83 |
| 10 | US20200082041A1 | Estimating physical parameters of a physical system based on a spatial-temporal emulator | 76 |
Citation counts favour older records simply by having had more time to accumulate them — treat them as a signal of influence within this corpus, not of current technical importance.
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 Model Compression for Deployment Targets 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 →Analysis
What the filing pattern signals
Reading volume, class mix and momentum together separates where claim space is genuinely occupied from where it only looks that way.
Concentration
24.0%
of 533 records held by top 5
A cluster, not a single blocker
The leading assignee holds 41 records and the next four bring the group to 128, or 24.0% of all 533 records in scope. That is meaningful concentration but well short of dominance — the remaining three-quarters of the field sits with a long tail across the other ranked assignees.
Assignee ranking, 100 companies
Momentum
-21%
2021→2024 filing change
A plateau, read carefully
Complete-year filings fell from 77 in 2021 to 61 in 2024. Because publication lags filing by roughly 18 months, the 2025 peak of 108 and the 19 filed so far in 2026 are both still under-counted, so this should not be read as declining interest.
Filing trend, 2017-2026
Technology mix
73.4%
of 533 records classed G06N
Claim density sits in the AI-model class
Three-quarters of records carry a G06N classification, confirming most compression inventions are framed as core model architecture. Thinner branches — speech/audio, transmission, business-process — show far less filed density, which is where claim room still exists.
IPC composition, 8 subclasses
Reach
272
US-filed records
US and PCT dominate receiving offices
The United States (272) and WIPO/PCT (79) together account for the bulk of filings, with Europe (64), India (42), China (32) and Canada (9) trailing. Filing strategy in this field still runs primarily through US prosecution with PCT as the secondary route.
Receiving offices, 6 jurisdictions
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 model compression for deployment targets, with the prior art for and against each one.
Source: Patsnap Eureka. Co-assignee relationships and derived observations. Derived from a Patsnap search on Model Compression for Deployment Targets 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
Assignee Landscape
Who is filing, and how actively
The ranked list covers 100 assignees across the 533 records in scope. Filing concentration sits at the top but recent-year momentum tells a more mixed story than cumulative counts alone.
Leader
41
records held
A clear top filer, but not filing recently
The top-ranked assignee holds 41 of the 533 records in scope, well ahead of fifth place at 18 and tenth place at 13. Momentum data shows this leader recording zero filings in the latest year, so cumulative claim strength and current activity are not the same signal.
Assignee ranking, position 1 of 100
Top 5
24.0%
of all 533 records
Meaningful but not overwhelming concentration
The five most active assignees combined hold 128 records, a quarter of everything in scope. The remaining three-quarters spreads across a long tail of the other 95 ranked assignees, most with single-digit record counts.
Top 5 combined, 128 records
Co-filing
5
co-assignee pairs recorded
Joint invention is rare but present
Only five co-assignee pairs appear in the dataset, and the strongest of them link one major filer to individually named inventors rather than to a second corporate assignee. Collaborative filing is the exception here, not the norm.
Co-assignee pairs, strongest links
Recent-year filing momentum by assignee
| Assignee | Recent year | YoY |
|---|---|---|
| Google LLC | 2 | -33% |
| DeepX Co., Ltd. | 1 | -83% |
| Samsung Electronics Co., Ltd. | 0 | -100% |
| X Development LLC | 0 | -100% |
| Intel Corporation | 0 | -100% |
| Huawei Technologies Co., Ltd. | 0 | — |
| Qualcomm Incorporated | 0 | -100% |
| L'Oréal SA | 0 | — |
Source: Patsnap Eureka. Assignee-level momentum. Derived from a Patsnap search on Model Compression for Deployment Targets 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
Next Steps
Where to take this analysis next
The figures above set the boundaries of the field. Turning them into a filing or freedom-to-operate decision means going record by record.
Check claim scope against the leading assignees
Cumulative record counts do not tell you what is actually claimed. Reading the independent claims of the top five assignees' most recent grants is the next step before drafting in a crowded class.
Explore assignee claims in Eureka →Trace the thin IPC branches for open claim room
Speech/audio and transmission-constrained compression show far lower filing density than the core AI-model class. Confirming that gap holds at the claim level, not just the classification level, is worth a closer pass.
Map white space in Eureka →Watch momentum, not just cumulative rank
Several major holders show sharp negative year-on-year filing swings even while holding large cumulative counts. Tracking who is still actively filing changes who to watch going forward.
Track assignee momentum in Eureka →Source: Patsnap Eureka. Forward-looking reading of the same dataset. Derived from a Patsnap search on Model Compression for Deployment Targets 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
Frequently asked questions
The ranking covers 100 assignees drawn from the 533 records in scope, and it is concentrated at the top: the leading assignee alone accounts for 41 records, and the top five combined hold 128 records, or 24.0% of all 533. That leaves a long tail of single- and few-filing entrants below the ranked leaders, so a newcomer is not necessarily competing against one dominant blocker but against a cluster of five to ten active filers.
Filing volume rose steadily from 2 records in 2017 to a peak of 108 in 2025, but the most recent one to two years always understate real activity because publication lags filing by roughly 18 months. The only clean complete-year comparison is 2021 to 2024, which shows 77 records falling to 61, a -21% change best read as the field settling into a mature filing plateau rather than losing momentum, since the underlying invention work continues to surface in later publications.
WO2024127462A1, filed by NEC Corporation on 2024-06-20, claims an automatic machine-learning pipeline that runs a neural architecture search, narrows the candidate search space with an alpha-analyzer, compresses the resulting small model by replacing sublayers within that narrowed space, and then deploys the lightweight model. It blocks that specific sequencing of search-then-narrow-then-compress, but pipelines that narrow the search space using a different signal, or that compress without that intermediate narrowing step, sit outside its claim language.
The thinnest technical branches by IPC share are speech and audio processing (G10L, 4.9% of the 533 records), digital transmission (H04L, 3.8%) and business-process-linked compression (G06Q, 4.3%), all well behind the dominant AI-model class (G06N, 73.4%). Claims that tie a compression decision to an audio codec constraint or a transmission-link budget, rather than a generic compute constraint, currently have far less filed prior art to clear.
Neural architecture search under an explicit hardware or latency constraint is the most heavily filed route and therefore the most contested, sitting inside the 73.4% of records classed under G06N. A new filer is generally better served by combining an underused constraint — an audio codec or transmission-link limit rather than a pure compute budget — with a compression mechanism, since that combination avoids direct overlap with the leading assignees' core claims while still building on validated deployment-focused claim language.
Answers are grounded in the same dataset. Derived from a Patsnap search on Model Compression for Deployment Targets 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 Model Compression for Deployment Targets in depth with Eureka
Go past this page: query the whole model compression for deployment targets 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.