How Projection Neural Networks Speed Up AI Predictions
A method for making artificial intelligence models faster and more efficient by using fixed, non-trainable projections to simplify complex data before processing.
Original patent title: “Projection neural networks”
A method for making artificial intelligence models faster and more efficient by using fixed, non-trainable projections to simplify complex data before processing. Granted to Google LLC in 2023 with 23 claims and 1 forward citation.
Key facts
Coverage
What does this patent actually cover?
This patent describes a way to build neural networks that are computationally cheaper to run. Instead of training every single part of the network, the system uses 'projection layers' that transform input data into a simpler, lower-dimensional format using fixed parameters that do not change during training. These fixed projections act like a filter, mapping complex inputs to a finite set of values (like 0 or 1) before the rest of the network processes them. By keeping these projection parameters constant, the system reduces the amount of math required to generate a final prediction, making the model faster and less memory-intensive.
The gap
What does this patent NOT cover?
- Does not cover neural networks where all parameters are updated during the training process.
- Does not cover models that do not use a finite set of values (like binary 0 or 1) for the projection function output.
- Does not cover systems that lack a projection layer as defined by the specific dot-product and thresholding mechanism described.
- Does not cover traditional deep learning architectures that rely solely on standard backpropagation for all layer weights.
These exclusions are unique to PatentBrief — derived from the actual claim language, not patent-office boilerplate.
What made this novel
The innovation is using fixed, non-trainable weights for the initial projection step. By freezing these parameters, the network avoids the computational cost of updating them, effectively 'compressing' the input data into a manageable format without losing the essential features needed for a prediction.
Schematic visualization of the patent's claim structure. Hand-drawn diagrams in progress for each landmark patent.
Where you've seen this
Real-world examples
On-device mobile text classification
Efficient keyword spotting in voice assistants
Real-time recommendation engines on low-power hardware
Why it matters
The bigger picture
As AI models grow in size, they become difficult to run on devices with limited power, like smartphones or IoT sensors. This patent provides a blueprint for 'lightweight' AI that can perform complex tasks without needing massive server-side computing power. It represents a shift toward efficiency-focused architecture design in machine learning.
Filed
July 13, 2020
Granted
January 3, 2023
Market context
Who's building on this
Companies in this space
Google is the primary developer of this technology, integrating it into their mobile and cloud-based machine learning frameworks. Other companies focused on 'TinyML' or edge computing, such as those developing specialized AI chips for phones and wearables, are exploring similar methods to reduce the computational footprint of neural networks.
Market impact
This technology supports the industry-wide push to move AI processing from the cloud to the 'edge' (local devices). By enabling smaller, faster models, it helps companies reduce latency and protect user privacy by keeping data processing on the device rather than sending it to a server.
Claim 1 — Plain English
What this patent covers
This patent describes a way to build neural networks that are computationally cheaper to run. Instead of training every single part of the network, the system uses 'projection layers' that transform input data into a simpler, lower-dimensional format using fixed parameters that do not change during training. These fixed projections act like a filter, mapping complex inputs to a finite set of values (like 0 or 1) before the rest of the network processes them. By keeping these projection parameters constant, the system reduces the amount of math required to generate a final prediction, making the model faster and less memory-intensive.
The clever bit
The innovation is using fixed, non-trainable weights for the initial projection step. By freezing these parameters, the network avoids the computational cost of updating them, effectively 'compressing' the input data into a manageable format without losing the essential features needed for a prediction.
What it does not cover
- Does not cover neural networks where all parameters are updated during the training process.
- Does not cover models that do not use a finite set of values (like binary 0 or 1) for the projection function output.
- Does not cover systems that lack a projection layer as defined by the specific dot-product and thresholding mechanism described.
- Does not cover traditional deep learning architectures that rely solely on standard backpropagation for all layer weights.
Patent timeline
Application submitted to the patent office
Application published, typically 18 months after filing
Patent officially issued
PatentBrief Score
Impact Score
Strong
Citation count
6/40
Early citations
Claim breadth
15/20
Broad claimsclaimsThe numbered statements at the end of a patent that legally define what the inventor owns.Read more →
Recency
20/20
Granted within 5 years
Assignee scale
20/20
Major company or institution
PatentBrief Impact Score — based on citation count, claim breadth, recency, and assignee scale. Not a legal assessment.
Heuristic Value Estimate
What this patent might be worth
$75K – $240K
Midpoint $150K · 14.1 yr remaining · industry ×1.6
Heuristic only — blends forward/backward citation counts, claim scope, time remaining, litigation history, and CPC-derived industry baseline. Real valuations need a professional appraisal.
The original legal language
Original claims
23 claims as filed with the patent office.
Concepts involved
Citations
Patent lineage
Cite this patent
Ravi, S. (2023). How Projection Neural Networks Speed Up AI Predictions (U.S. Patent No. 11,544,573). U.S. Patent and Trademark Office. https://patentbrief.org/patent/us/11544573/llama-large-language-model-architecture
Auto-generated from the patent record. Double-check author order and the issue date against the official USPTO document before submitting.
Embed
Add this patent to your site
Drop this plain-English patent card into any blog post or article — free, no signup. It always links back to the full breakdown here.
<div data-patentlens-widget data-patent-number="US11544573"></div> <script src="https://patentbrief.org/embed.js" async></script>
Stay in the loop
Get a weekly digest of new patents.
One email per week. No spam. Unsubscribe anytime.
Keep exploring
Related patents you should know
US 4683195 · 1987
How to Make Billions of Copies of a DNA Segment
This patent describes the Polymerase Chain Reaction (PCR), a method to rapidly create many copies of a specific piece of DNA or RNA, enabling its detection and analysis.
Cetus Corp
US 8697359 · 2014
How to Edit Genes in Human Cells Using an Engineered CRISPR System
This patent describes an engineered CRISPR-Cas9 system for precisely cutting DNA in eukaryotic cells to change how genes work, opening the door for gene editing in complex organisms.
Massachusetts Institute of Technology
US 7657849 · 2010
How the iPhone's Slide-to-Unlock Gesture Works
Apple's 2010 patent describes unlocking a device by dragging a specific graphical image across the touchscreen along a predefined path, a gesture that became iconic with the original iPhone.
Apple Inc
US 4733665 · 1988
How Doctors Implant a Permanent Stent Using a Balloon
This patent describes the method for placing a permanent, expandable wire mesh tube inside a blood vessel or other body tube using a balloon-tipped catheter to widen it and keep it open.
Expandable Grafts Partnership
US 4965188 · 1990
How to Make Many Copies of a DNA Piece with Heat
This patent describes the Polymerase Chain Reaction (PCR) method, a technique to make millions of copies of a specific DNA segment using a heat-resistant enzyme and repeated temperature changes.
Cetus Corp
US 4235871 · 1980
How to Encapsulate Active Materials in Lipid Bubbles Efficiently
This patent describes a method for trapping biologically active substances inside tiny, multi-layered fat bubbles called liposomes, using a specific water-in-oil emulsion and gel-forming process to improve how much material gets captured.
Individual
More to explore
More in AI & Machine Learning
US 10452978 · 2019 · Google LLC
How AI Models Understand Language Using 'Attention'
US 6523026 · 2003 · Huntsman International LLC
How Computers Find Hidden Connections Between Different Fields of Knowledge
US 11615208 · 2023 · Capital One Services LLC
How Cloud Systems Automatically Create and Train AI Data Models
US 10402750 · 2019 · Facebook Inc
How Facebook Uses Deep Learning to Predict What You Might Like
New to patents?
Common Questions
Frequently Asked Questions
What does How Projection Neural Networks Speed Up AI Predictions cover?
A method for making artificial intelligence models faster and more efficient by using fixed, non-trainable projections to simplify complex data before processing.
Who owns patent US 11544573?
Google LLC owns this patent, granted in 2023.
When does this patent expire?
This patent is expected to expire on January 3, 2043, when the invention enters the public domain.
What is patent US 11544573 cited by?
This patent has been cited by 1 later patents that build on its ideas.
What problem does this patent solve?
As AI models grow in size, they become difficult to run on devices with limited power, like smartphones or IoT sensors. This patent provides a blueprint for 'lightweight' AI that can perform complex tasks without needing massive server-side computing power. It represents a shift toward efficiency-focused architecture design in machine learning.
What does this patent NOT cover?
Does not cover neural networks where all parameters are updated during the training process.
Same assignee
More from Google LLC
Patent monitoring